0% found this document useful (0 votes)

1K views

Interactive Computer Graphics Lesson 16

The purpose of these notes is to provide students with a basic knowledge of computer graphics. The chapters are arranged in an order that progresses from devices to software. Computer graphics and computer programming are inseparable.

Uploaded by

Rizwan Ali

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

1K views

Interactive Computer Graphics Lesson 16

Uploaded by

Rizwan Ali

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 302

Interactive Computer Graphics

ME 573 - CS 535 Lecture Notes D. C. Anderson August, 2001

Interactive Computer Graphics

ME 573 - CS 535 Lecture Notes

D. C. Anderson

August, 2001

Copyright 1988-2001 D. C. Anderson

Preface
The field of computer graphics has experienced a number of hardware and software revolutions since the early 1970s when these notes began. The purpose of these notes now, however, are very much the same as then, to provide students with a basic knowledge of, and practical experience with, the fundamental mathematics, algorithms, representations and techniques that are fundamental to interactive computer graphics applications. The chapters are arranged in an order that progresses from devices to software, from two dimensions (2D) to three dimensions (3D), and from simple geometric forms to the more complex curves, surfaces and solids. At times, this causes some topics to be re-visited, such as clipping and transformations in 2D and 3D. This order was chosen to facilitate a more natural topical progression for learning. Students accumulate knowledge and experience with relatively simpler topics first, such as 2D programming techniques, and then are ready to combine these with the later topics that are more mathematically and computationally challenging. Computer graphics and computer programming are inseparable. Typically in the course that uses these notes, four programming projects are assigned over sixteen weeks. Each project requires three to four weeks, coinciding as much as possible with the lecture topics. The goals of the projects are to give the student in-depth practice with the concepts while designing, coding and debugging a relevant and non-trivial application. Typically, the project topics are (1) basic 2D display with minor interaction, (2) fully interactive 2D drawing requiring dynamic data structures, (3) the wireframe 3D pipeline with projection, 3D clipping, and hierarchical transformations with little interaction, (4) 3D hidden surface removal and shading, such as scan-line rendering or ray tracing. Partially incomplete source code for the first project usually is given to the students to demonstrate basic graphical programming techniques and styles. This has proven to be effective in speeding the learning of programming details as well as various operating system and graphics package minutia. This also paves the way for later concentration on mathematical and algorithmic concepts. Over the past 25 years, there have been four generations of graphics devices used for the projects, starting with the Imlac minicomputer vector display system, progressing through Megatek 3D vector systems, raster graphics workstations and now personal computers. Each

new display technology and new generation of system software spawned a new generation of graphics programming package. Such is the case for the GRAFIC package which was developed during the first days of X Windows (version 10) to support the teaching of computer graphics. GRAFIC was designed to provide basic, multi-language (initially C, Pascal and Fortran; and now C++ as well), multi-platform access to window-based raster graphics. An important objective of GRAFIC was that its use should require little training in the package itself. Now, versions of GRAFIC exist for X Windows for Unix workstations, IBM-compatible personal computers with Microsoft Windows , and Apple Macintosh computers. GRAFIC is available as-is and free of charge from the internet site shown below. There are many individuals who have contributed to these notes over the years. Michael Bailey taught the course Interactive Computer Graphics with the author for several years, established some of the early topics, and developed some of the early versions of the written notes. Warren Waggenspack contributed significantly to the presentation of curves and surfaces while at Purdue, and has been a valued collaborator who has encouraged continued work on the notes by the author. Philip Cunningham helped with the presentation of the reflectance and shading models. Joseph Cychosz contributed to the presentation on color and some of the ray casting algorithms. Gregory Allgood implemented the first versions of X Windows GRAFIC. To these individuals, the author extends sincere appreciation for their invaluable help.

David C. Anderson Purdue University email: [email protected] internet: http://www.cadlab.ecn.purdue.edu/

iii

Bibliography
The following is a partial list of texts covering various topics in computer graphics. They are categorized by their primary emphasis and alphabetically by first author. General Computer Graphics Texts: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. Berger, M., Computer Graphics with Pascal, The Benjamin/Cummings Publishing Company, Inc., 1986. Dewey, B. R., Computer Graphics for Engineers, Harper & Row, 1988. Foley, J.D. and Van Dam, A., Fundamentals of Interactive Computer Graphics, AddisonWesley, 1982. Giloi, W.K., Interactive Computer Graphics, Prentice-Hall, Inc, 1978. Harrington, S., Computer Graphics - A Programming Approach, McGraw-Hill Book Company, 1987. Hearn, D., Baker, M. P., Computer Graphics, Prentice-Hall, Inc., 1986. Hill, F. S. Jr., Computer Graphics, Macmillan Publishing Company, 1990. Newman, W.M. and Sproull, R.F., Principles of Interactive Computer Graphics, Second Edition, McGraw-Hill, 1979. Plastok, R. A., Kalley, G., Theory and Problems of Computer Graphics, Shaums Outline Series in Computers, McGraw-Hill Book Company, 1986. Pokorny, C. K., Curtis, F. G., Computer Graphics: The Principles Behind the Art and Science, Franklin, Beedle & Associates, 1989. Rogers, D.F., Procedural Elements for Computer Graphics, McGraw-Hill Book Company, New York, 1985.

Geometric Modeling Emphasis: 12. 13. 14. Barnsley, M. F., Devaney, R. L., Mandelbrot, B. B., Peitgen, H.-O., Saupe, D., and Voss, R. F., The Science of Fractal Images, Springer-Verlag, 1988. Ding, Q., Davies, B. J., Surface Engineering Geometry for Computer-Aided Design and Manufacture, John Wiley & Sons, 1987. Farin, G., Curves and Surfaces for Computer Aided Geometric Design - A Practical Guide, Academic Press, Inc., Harcourt Brace Jovanovich Publishers, 1988.

15. 16. 17. 18. 19. 20. 21.

Hoffmann, C. M., Geometric and Solid Modeling - An Introduction, Morgan Kaufmann Publishers, Inc. San Mateo, California, 1989. Lyche, T., Schumaker, L. L., Mathematical Methods in Computer Aided Geometric Design, Academic Press, Inc., Harcourt Brace Jovanovich Publishers, 1989. Mantyla, M., An Introduction to Solid Modeling, Computer Science Press, Inc., 1988. Mortenson, M. E., Geometric Modeling, John Wiley & Sons, New York, 1980. Mortenson, M. E., Computer Graphics - An Introduction to the Mathematics and Geometry, Industrial Press, Inc., 1989. Rogers, D.F and Adams, J.A., Mathematical Elements for Computer Graphics, McGrawHill, 1976. Yamaguchi, F., Curves and Surfaces in Computer Aided Geometric Design, Springer-Verlag, 1988.

Realistic Image Generation Emphasis: 22. 23. 24. 25. 26. 27. 28. Deken, Joseph, Computer Images, Stewart, Tabori, & Chang, New York, 1983. Glassner, A. S. (ed.), An Introduction to Ray Tracing, Academic Press, Inc., Harcourt Brace Jovanovich Publishers, 1990. Greenberg, D., Marcus, A., Schmidt, A., Gorter, V., The Computer Image: Applications of Computer Graphics, Addison-Wesley, 1982. Hall, R., Illumination and Color in Computer Generated Imagery, Springer-Verlag, 1989. Mandelbrot, Benoit, The Fractal Geometry of Nature, W.H. Freeman and Company, New York, 1983. Overheim, R.D. and Wagner, D.L., Light and Color, John Wiley & Sons, 1982. Schachter, B., Computer Image Generation, John Wiley & Sons, New York, 1983.

Contents
Chapter 1. Introduction .......................................................................................................... 1.1 Historical Perspective ....................................................................................... 1.2 Notes Overview................................................................................................. 1.3 Notation............................................................................................................. Chapter 2. Graphics Devices ................................................................................................. 2.1 Static Graphics Devices .................................................................................... 2.1.1 Plotters................................................................................................ 2.1.2 Electrostatic Plotters........................................................................... 2.1.3 Laser Printers ..................................................................................... 2.1.4 Film Recorders ................................................................................... 2.2 Dynamic Graphics Devices............................................................................... 2.2.1 The Cathode Ray Tube (CRT) ........................................................... 2.2.2 The Point Plotting Display ................................................................. 2.3 Raster Displays ................................................................................................. 2.3.1 Video Technology .............................................................................. 2.3.2 The Bitmap Display ........................................................................... 2.3.3 Pixel Memory Logic .......................................................................... 2.3.4 Gray Scale Displays ........................................................................... Chapter 3. Color..................................................................................................................... 3.1 Light .................................................................................................................. 3.2 Color Perception ............................................................................................... 3.3 Color CRT Operation........................................................................................ 3.4 Additive and Subtractive Color......................................................................... 3.5 Color Representations ....................................................................................... 3.5.1 RGB Color Space ............................................................................... 3.5.2 CMY and CMYK Color Spaces......................................................... 3.5.3 YIQ Color Space ................................................................................ 3.5.4 XYZ Color Space ............................................................................... 3.5.5 HSI Color Space................................................................................. 3.5.6 HSV Color Space ............................................................................... 3.5.7 Luv Color Space................................................................................. 3.5.8 Lab Color Space ................................................................................. 3.6 References ......................................................................................................... 3.7 Review Questions ............................................................................................. Chapter 4. Basic Graphics Software ...................................................................................... 4.1 Hardcopy Plotting ............................................................................................. 4.2 Drawing Lines................................................................................................... 4.2.1 Bresenham Line Stepping Algorithm................................................. 4.2.2 The Plot Routine ................................................................................ 4.3 Characters.......................................................................................................... 4.3.1 Character Definitions ......................................................................... 1 1 2 2 4 4 4 8 10 10 12 12 12 14 14 15 16 17 19 19 21 22 25 26 27 28 29 30 32 36 37 38 40 41 42 42 44 45 47 49 49

4.4

4.5 4.6

4.3.2 Font Data Structures........................................................................... Raster Software ................................................................................................. 4.4.1 Rasterization....................................................................................... 4.4.2 Raster Display Drawing Operations................................................... References ......................................................................................................... Review Questions .............................................................................................

51 53 54 55 58 59 61 61 62 64 66 67 70 70 70 72 73 73 75 77 78 79 82 84 84 86 86 88 90 91 92 93 93 95 97 98 98 101 102 104 105

3 6

4 5

in-line pen carriage

Figure 2-3. Multiple Pen Carriages. chamber is held inside a solenoid coil. When activated by the plotter electronics, the solenoid lifts the pen above the plotting surface. X and Y motions are controlled by stepping motors. The solenoid and motors are controlled by a computer. Only one step is taken at a time. The resolution of a plotter is a measure of its accuracy, or step-size. Typical resolutions are 0.002 and 0.005 inch per step, or 500 or 200 steps per inch, respectively. There are several critical factors in the design of plotters: speed The speed of a plotter is typically given in steps per second. The time required to complete a given drawing is a function of the distance travelled by the pen, and the resolution and speed of the plotter. For example, consider drawing an 8 by 10 box around a page using a plotter with 500 steps per inch and a speed of 500 steps per second. The time to complete the drawing is computed as follows: time (sec) = distance (inch) / {resolution (inch/step) * speed (step/sec)} or, time = (8+8+10+10) / {0.002 * 500} time = 36 sec (Note the plotter travels at 1 inch/second.) accuracy Generally, increasing accuracy requires slower speeds and more massive structures to maintain mechanical rigidity. ink flow High speed plotters must force ink from the pen because gravity alone is insufficient. Configurations There are three basic configurations for connecting a plotter to a computer: off-line, online, and spooled (buffered). Off-line configuration means the plotter is not physically connected to the computer, so data must be manually transferred. The plotter reads plotting information from

Chapter 2. Graphics Devices

2.1 Static Graphics Devices

a tape that must be loaded and unloaded periodically. The plotting programs write this information on the tape and, at a designated time, the tape is removed and queued for the plotter.

Computer

tape

Plotter

Manually Carry Tape Figure 2-4. Off-line Plotter Configuration. On-line configuration means there is a direct data communication interface between the computer and the plotter. As the plotting program executes in the computer, calls to plotting routines cause plot commands to be sent to the plotter. The plot commands are executed by the plotter hardware, i.e. the pen is moved. This is a typical configuration for a personal computer and a plotter.

Computer

plot commands

Plotter

Figure 2-5. On-line Plotter Configuration. It usually does not make sense to configure a plotter on-line with a multi-user, time-shared computer system because several plotting programs can be executing at once. In this case, it is necessary to couple the two previous approaches, on-line and off-line, in a manner that allows several plotting programs to create plot data, yet send only one programs data at a time to the plotter. Spooling is a method for accomplishing this. Spooling Current time-shared systems perform a combination of off-line and on-line plotting, called spooling. This is similar to how line printers are operated. Storing and plotting can occur simultaneously (with proper provisions for simultaneous access to files). Plotting programs create intermediate plot files when they execute. When a program finishes, the plot file is queued for plotting, that is, transferred onto a disk area set aside for plot files waiting to be plotted. This intermediate disk storage is a buffer between the programs creating plot data and the slower plotter,

Chapter 2. Graphics Devices

2.1 Static Graphics Devices

which can only process one plot at a time. Another computer program, sometimes executing in another computer, accesses the queued plot files periodically and sends them to the plotter. The plotting programs are off-line with respect to the plotter, and the intermediate program is on-line. Disk Buffer store

plot

Computer

Spooling Processor Figure 2-6. Spooling Configuration.

Plotter

2.1.2 Electrostatic Plotters

Pen plotters are relatively slow devices. Electrostatic plotting was developed as a compromise between speed and accuracy. Electrostatic plotting is a process that, in effect, charges small dots on a piece of paper and then immerses the paper with toner, dry powdered ink. The ink adheres to charged dots on the paper and falls off the rest (Figure 2-7). Dry and Remove Electrode

Toner Applicator

Writing Head Side View

Paper Roll

epp

Paper Motion

Top View of Writing Head (magnified) Figure 2-7. Elements of an Electrostatic Plotter.

Chapter 2. Graphics Devices

2.1 Static Graphics Devices

One advantage of electrostatic plotting is that the printing process is almost entirely electronic, so the problems of speed and accuracy that limit mechanical pen plotters are eliminated. The drawback is that the resulting image must be drawn as an array of dots. Each dot is called an electrostatic plotter point or an epp. The plotting process involves the computer sending data to the plotter for each line of epps. The data for a line of epps is a series of binary numbers, 1or 0, representing dot (on) or no dot (off) for each writing head epp position. The computer must compute the binary values for all epps based on the image that is to be drawn. After a complete line of epps is received, the plotter controls the writing head to charge the paper at epps whose values are on. As the paper advances, new lines are charged and the charged lines reach the toner applicator where the ink is applied. The process continues down the entire page. The page cannot back up, so all plotting must be done in one pass. The density of epps, or resolution, is critical to the quality of the image. Typical electrostatic plotters have resolutions of 100 to 250 epps per inch. One hardware modification to increase resolution is to arrange the writing head in two rows of epps, one for even epps and another for odd. A single line of epps is printed on the paper in two steps that are synchronized in time so that the even and odd epps align along a line. t

Figure 2-8. Staggered Epp Writing Head Operation. Some past and present electrostatic plotter manufacturers are Versatec, Houston Instruments, Varian, and Gould. Prices range from $5,000 to $50,000, widths vary from 8 inches to 6 feet, speeds vary between 1 to over 6 inch/sec., and resolutions vary from 80 to 250 epps per inch. Color printing can be done with multiple passes over the same paper, once for each color component. This will be described later in the chapter on color. Electrostatic plotters have been replaced by the more popular laser printers.

Chapter 2. Graphics Devices

2.1 Static Graphics Devices

2.1.3 Laser Printers

Laser printers operate like electrostatic printers, except the printing engine is a laser beam instead of a writing head. The computer sends digital information (commands) to the controller, which carefully controls the intensity and direction of the laser beam through a rotating polygon mirror. The mirror deflects the beam to the proper location (dot) on the photoreceptor drum. One line of dots along the drum is written (charged) at a time. The drum rotates, imparting the charge onto the paper passing under it. The toner then adheres to the charged locations on the paper. Laser printers are capable of much higher resolutions than electrostatic plotters, from 300 to thousands of dots per inch (dpi). Polygon mirror Computer Controller Laser Laser Beam Photoreceptor drum Toner cartridge Paper Fuser

Figure 2-9. Elements of Laser Printers. Some laser printer engine manufacturers are Canon, Ricoh, Kyocera, and Linotype. Prices range from $1,000 to over $50,000 and speeds vary from a few to 20 or more pages per minute.

2.1.4 Film Recorders

These devices write on film. The image is drawn by drawing three monochrome (one color) images, each through the appropriate filter. Another variety of film recorder is used in conjunction with a raster display. The process is

Chapter 2. Graphics Devices

2.1 Static Graphics Devices

Cathode Color Wheel Ray Tube Green computer video output Red Blue

Camera Color-filtered image recorded on film

Figure 2-10. Elements of a Film Recorder. similar to the film recorder above, however the image is not computed for the recorder, but is instead captured from a screen display. Color Wheel Switch
B/W CRT

Camera Color-filtered image recorded on film.

Color Display

R G B

CRT

Figure 2-11. Elements of a CRT Film Hardcopy Recorder.

Chapter 2. Graphics Devices

2.2 Dynamic Graphics Devices

Dynamic graphics describes the use of refresh displays for graphical interaction which can involve rapidly changing images and motion.

2.2.1 The Cathode Ray Tube (CRT)

At the heart of all dynamic graphics systems is the cathode ray tube (CRT), the fundamental electronics for creating visible images on a screen. The basic elements of the CRT are shown in Figure 2-12. The beam starts as a flood of electrons generated in the filament, cathode X & Y deflection systems beam focusing system screen with phosphor coating

filament cathode control grid accelerating plates Figure 2-12. Elements of a CRT. and control grid. The flood of electrons is focused into a concentrated beam by the focusing system and accelerated into a narrow beam of high energy electrons by the accelerating plates. The beam is deflected by charges in two pairs of charged plates, one horizontal and one vertical, in the deflection system. The horizontal and vertical deflection voltages are controlled by signals from the computer.

2.2.2 The Point Plotting Display

The basic display process begins with the computer setting the voltages in the horizontal and vertical deflection plates to appropriate values for the desired beam location. This is done

Chapter 2. Graphics Devices

2.2 Dynamic Graphics Devices

through digital-to-analog (D/A) converters connected between the computer and the display. The computer sends digital values to the D/A converters that convert the values to voltages. The output voltages of the D/A converters are connected to the X and Y deflection plates. The computer then pulses the beam intensity to a pre-set maximum value. Where the beam strikes the screen, the phosphor glows, causing a visible dot. The beam must be turned off quickly to avoid burning the phosphor. This forms the basic point plotting display (Figure 2-13). X Computer Intensity Y D/A D/A D/A CRT

Figure 2-13. Configuration of a Point Plotting Display. When the beam is turned off, the glow decays rapidly based on the persistence, or rate of light decay, of the screen phosphor. The glow decays very quickly, so the image must be refreshed Intensity 100% Threshold Time Figure 2-14. Decay in Intensity of Phosphor Light. (re-drawn) often, typically every 1/30 - 1/60 sec. (30 - 60 cycles/sec or hertz), even if the picture does not change. If the glow fades too much before it is refreshed, the eye will detect this intensity Intensity Start End Average intensity Time Refresh Cycle Figure 2-15. Intensity Variations Due to Refresh Cycles.

Chapter 2. Graphics Devices

2.3 Raster Displays

variation as flicker. Hardware can be added to the basic point plotting display to produce two different types of refresh display technology: Vector continuous random lines: the beam motion and intensity are controlled by a vector generator that draws line as one continuous beam motion, analogous to pen plotting. Display commands from the computer direct the drawing of the picture. Raster ordered dots: the beam motion always follows the same path. The picture depends on the appearance of the dots, analogous to laser printing. The computer draws the picture by setting values in computer memory. We will focus on raster technology first.

2.3 Raster Displays

The common usage of the phrase raster graphics means the image is drawn in dots, that can be black & white, in shades of gray, or in colors. Raster graphics is based on the television CRT scanning display technology, or video.

2.3.1 Video Technology

A typical television has about 525 scan-lines (rows of dots across the screen) and 640 dots per scan-line. The signal received by a television defines every dot on the screen during every scancycle. The dots are called picture elements, or pixels. The length of the television scan-cycle is 1/30 second. In 1/30 second, the beam visits (displays) all pixels, illuminating them with the correct color/intensity. Televisions actually display the image using interlace. The even-numbered scan-lines are refreshed during one 1/60 second cycle, and the odd-numbered lines are refreshed during the next 1/60 second. Interlace smooths the displayed image by reducing the effects of beam motion variations and phosphor decay. Now consider creating the video signal from a computer instead of a camera and transmitter. Such a raster display must draw each pixel 30 times per second to maintain the image. Just as a television constantly receives new images from a transmitter, the raster display hardware

Chapter 2. Graphics Devices

2.3 Raster Displays

Figure 2-16. Beam Motion for Interlaced Raster Display. must repeatedly process its data to the display screen to refresh the image. We call this hardware a raster display processor.

Computer Program

Computer Data (image)

Raster Display Processor Every Refresh Cycle (30 to 60 hz)

CRT

When the Picture Changes

Figure 2-17. Raster Display Process.

2.3.2 The Bitmap Display

The simplest raster display processor is a bitmap display, or a black & white (B/W) display. It is so named because it uses one memory bit to determine the on/off state of a particular pixel drawn by the display processor. The major difference between the bitmap display and the electrostatic plotter is that it is no longer a static display device. Pixels can be turned on or off each refresh cycle. Digital data corresponding to the screen image must be maintained for the display processor. This data is stored in computer memory, called pixel memory. In the case of a bitmap display, pixel memory contains one bit (binary 0 or 1) for each pixel. Another term for pixel memory is frame buffer. We refer to the numeric value of each pixel as pixel value. A bitmap display has pixel values that are 0 or 1, corresponding to black or white (or vice-versa).

Chapter 2. Graphics Devices

2.3 Raster Displays

Pixel Memory row v pixel h,v

0,0

CRT Screen text

column h

Figure 2-18. Pixel Memory Organization.

2.3.3 Pixel Memory Logic

Pixel memory can be considered as an array of numbers whose dimensions depend on the screen size. The memory is organized into height scan-lines, each containing width pixels. Think of this as a two dimensional array, pixels[width,height] (Figure 2-19). We impose (0,0) height scan-lines width pixels per scan-line h

pixels[h,v]

v Figure 2-19. Pixels Array Variables. a coordinate system on the pixel array by numbering each pixel by its position as the hth pixel in the vth scan-line, i.e. the coordinate (h,v), with the first pixel in the upper-left corner having coordinates (0,0). Historically, the video display order dictates this choice of coordinate system. It is now possible to alter the display by changing the values in the pixels array. For

Chapter 2. Graphics Devices

2.3 Raster Displays

example, setting a pixel is simply the statement: pixels[h,v] = BLACK. This pixel will be displayed black by the display processor during the next refresh cycle. The statement pixels[h,v] = WHITE clears the pixel, i.e. sets it to white, which is often the background color. One could think of this as erasing. A 512 by 512 black-and-white display requires 256K (K = 1024) bits of memory. Also, every pixel is displayed each refresh cycle, even if you have not drawn anything on the screen. That is, a pixel always has a value that is displayed.

2.3.4 Gray Scale Displays

In order to display shades of gray the display processor hardware must allow a range of intensity values to be displayed at each pixel location, called a gray scale display. Instead of the two values on and off, pixel values are now integers that will be converted into intensities by the hardware each refresh cycle. The pixels array contains integers with sufficient bits to hold the range of pixel values. Suppose we need 1024 gray levels. This means that pixel memory must hold pixel values between 0 and 1023, so at least 10 bits per pixel are needed (1024 possibilities = 2 ). The resulting pixel memory can be envisioned as a two dimensional array of 10 bit integers, i.e. int pixels[512,512], that can be conceptualized as a three dimensional screen as shown in Figure 2-20. Pixel memory is now 512 by 512 by 10 bits deep , requiring 2,621,440 bits.
10

Chapter 2. Graphics Devices

2.3 Raster Displays

bits per pixel (0,0) h pixel[h,v]

v Figure 2-20. Three-Dimensional Conceptual View of Screen Pixel Memory.

Chapter 2. Graphics Devices

3.1 Light

Chapter 3. Color
The perception of color is a complex problem that has been studied for hundreds of years. Creating a meaningful color display requires an understanding of the various physical and perceptual processes, including a fundamental understanding of the physical properties of light, human color perception, color CRT operation and computer models for color representation.

3.1 Light
Light is electromagnetic radiation, energy, that is emitted from light sources such as incandescent bulbs and the sun, reflected from the surfaces of objects, transmitted through objects, propagated through media such as air, and finally reaches our eyes. There are several models that have been proposed to describe the physical behavior of light. Generally, it is viewed as having both the properties of particles and waves. The model most appropriate for computer graphics describes light as an oscillating electromagnetic wave, as illustrated in Figure 3-1. As a wave, light
amplitude wavelength

propagation direction

Figure 3-1. An Electromagnetic Wave. has the wave properties of frequency , wavelength , and amplitude. It is known that frequency c and wavelength are related by = -- , where c is the speed of light in the medium (for a vacuum, c = 2.998 x 108 meters/sec). Wavelength is usually given in units of micrometers m (10-6) or nanometers nm (10-9). The electric field intensity of the wave varies sinusoidally in time and space as a function of amplitude and frequency. Visible light, the light that the human eye can perceive and thus the light that we can see,

Chapter 3. Color

3.1 Light

is only part of the entire electromagnetic spectrum, as shown in Figure 3-2 and Figure 3-3. The term
Visible X Rays Gamma Rays Ultraviolet Infrared Microwave

380 10-2 10-1 1

770 104 105 106 107

10 102 103 wavelength in nm

Figure 3-2. Electromagnetic Radiation. Yellow-Green Yellow Blue-Green

Orange

Violet

Green

Blue

350

400

450

500 550 600 wavelength in nm

650

Red 700

750

Figure 3-3. Colors in the Visible Light Spectrum. light for our purposes means electromagnetic radiation with wavelengths in the range 380 nm to 770 nm. Generally, light that we see is not just a single wavelength, but consists of a continuous, non-uniform distribution of single-wavelength (monochromatic) components. The graph of intensities versus wavelengths forms a spectral distribution, or spectral reflectance curve, as illustrated in Figure 3-4. The term spectral means variance with, or dependence upon, wavelength.

energy intensity

wavelength Figure 3-4. Spectral Curve of a Color of Light.

Chapter 3. Color

3.2 Color Perception

Humans perceive color by differentiating the chromatic and achromatic characteristics of light. The achromatic characteristics are largely determined by the intensity and the chromatic by the wavelength. The differentiation is the result of many factors: 1. the physical characteristics of the object itself and its capacity to reflect or absorb certain wavelengths of light, 2. 3. 4. 5. 6. 7. the properties of the light source illuminating the object, the medium through which the light travels and the distance through the medium, the properties of the surrounding objects or area, the biochemical state of the eye and visual system upon stimulation, the transmission characteristic of the receptor cells and neural centers, the subjects previous experience with the object or sensation.

The first four factors pertain to the physical properties of the objects and their surrounding environment, and can be precisely determined. The last three deal with the physiological characteristics of the human visual system and vary from person to person, even with normal visual systems. For example, individuals will discern different transition points at which an object appears either red or orange, even though the physical situation does not vary [Cychosz]. The human eye consists of a sensor array of photo-receptors known as the retina, and a lens system under muscle control that focuses the visual stimulation on the retina. The retina consists of two types of sensors: rods and cones. Rods are primarily used for night vision and respond achromatically to a given stimulus. Cones are used for color vision in normal lighting and respond primarily to wavelength. Color is interpreted by the brain based on the stimuli from three types of cones that are distributed over the retina. Each cone absorbs either reddish light, greenish light or bluish light. The distribution of cones is not uniform. Sixty-four percent of the cones absorb red pigment, thirty two percent absorb green, and only two percent absorb blue. Figure 3-5 shows the relative sensitivities of the three cone types [Hunt 1987, page 13]. The stimulus information sent to the brain through the optic nerve by the receptors conveys a luminance (also called lightness, brightness and intensity) and two color ratios: a red-to-green ratio and a yellow-to-blue ratio. As a result of experimental data, it is believed that the eye responds primarily to intensities

3.3 Color CRT Operation

Pixel Memory pixel h,v pixel value R

VLT

B to CRT guns

Figure 3-8. Mapped Color Display. from the VLT location addressed by the pixel value for the pixel color. Mapped color systems provide a means for reducing the size of pixel memory (and therefore the cost of a system) when it is acceptable to choose a subset of colors from a large palette of possible colors. For example, most workstations today with color displays are mapped systems capable of displaying 256 colors chosen from a palette of 224 possible colors. (The number 224 is over 16.6 million.) To understand the physical configuration of such a display, break down the quoted phrase in the previous sentence. 256 colors means simultaneous colors on the screen at one time, so pixel memory must be able to store 256 different colors. Therefore, pixel memory must have log2 256, or 8, bits per pixel (i.e. 28 = 256). Palette is a term often used for VLT, so 256 colors also means that there must be 256 VLT entries that can be indexed by the pixel values. All that remains is the format of each VLT entry. The phrase palette of 224 possible colors tells us that each entry must be able to store 224 colors, i.e. each entry must contain 24 bits. The entry must be divided into three components for RGB, so we can safely assume that each VLT entry contains 24 divided by 3, or 8 bits per component. It would be very unusual not to use the same number of bits for each component. Figure 3-9 shows a schematic diagram of the configuration of such a display system. A direct color system capable of displaying 224 colors would require 24 bits per pixel. This is a significantly more costly system than a mapped display with a palette of 224 colors. The difference is sometimes subtle: the mapped system can display only 256 colors at once because each pixel can represent only 256 possible values. Each pixel value is mapped to a VLT entry, however, that can contain 224 possible values (colors). The 24-bit direct system, sometimes called a full color system, can display 224 colors at once, but has three times as much pixel memory (24 bits per pixel versus 8 bits per pixel). The additional memory required for the VLT is generally

Chapter 3. Color

3.4 Additive and Subtractive Color

VLT (256 entries) Pixel Memory pixel h,v R

3.5.3 YIQ Color Space

In 1953, the NTSC (National Television Standards Committee) recommended a set of transmission primary colors, known as YIQ, as a basis for the broadcast color television signal [Smith]. This is basically an adjustment to the standard RGB space that was considered to be more appropriate for television transmission and viewing. Like many color models, it is related to the

Chapter 3. Color

3.5 Color Representations

RGB space through a linear transformation that is compactly represented in the matrix equation: Y 0.299 0.587 0.114 R = 0.596 0.275 0.321 G I Q 0.212 0.523 0.311 B The conversion from YIQ to RGB involves the inverse of the matrix: R 1 0.956 0.620 Y G = 1 0.272 0.647 I B 1 1.108 1.705 Q The Y coordinate is the luminance, the brightness perceived by the viewer, and is the only signal received by a black-and-white television. The I and Q components are two chrominance values that determine the color of a pixel. They are derived from the modulation technique used in NTSC broadcast signals.

3.5.4 XYZ Color Space

In 1931, the Commission Internationale de l Eclairage (CIE) defined three standard primary colors, called X, Y and Z, as alternatives to R, G and B for color matching. The XYZ coordinates are based on three spectral matching functions, X(), Y(), and Z(), shown in Figure 3-14, that were developed based on color matching experiments and desired mathematical
1.8 1.6 1.4

Z matching curve

Y matching curve
1.2 1 0.8 0.6 0.4 0.2 0 380

X matching curve

430

480

530

580

630

680

730

780

wavelength (nm)

Figure 3-14. CIE Spectral Matching Curves. properties. The Y() curve matches the luminance response curve of the human eye, i.e. the eyes sensitivity to spectral wavelengths. These color matching functions are tabulated at 1 nm

Chapter 3. Color

3.5 Color Representations

wavelength intervals [Wyszecki and Hall]. The CIE coordinates, (X,Y,Z), are computed from the spectral curve of a given color, I(), as integrals, X = k I()X() d Y = k I()Y() d Z = k I()Z() d

where k is a scaling constant appropriate for the lighting situation [Foley, p. 580]. In practice, I() is tabulated at 1 nm intervals and the integration is done numerically by summation. The XYZ coordinates include luminance information in addition to chroma. To remove the luminance factor, XYZ coordinates can be scaled to normalized values, i.e. each coordinate is divided by the sum X+Y+Z, resulting in xyz coordinates (lower case), called chromaticities. Note that x+y+z = 1 by definition, so only the x and y coordinates are needed. The two dimensional plot of these (x,y) values is called a chromaticity diagram and bounds the colors in the XYZ color space. The CIE xyz chromaticity diagram is shown in Figure 3-5.

y chromaticity

x chromaticity

Figure 3-15. CIE xyz Chromaticity Diagram. The chromaticities of a color can be measured with an incident light chromaticity meter. This can be done for a color monitor to calibrate its red, green, blue and white colors for precise color presentation. From the measured chromaticities it is possible to compute the transformation between RGB to XYZ. Mathematically, the chromaticities are the (x,y,z) values corresponding to the RGB primaries (red=(1,0,0), etc.). These reference colors are solved to form the transformation

Chapter 3. Color

3.5 Color Representations

of the form R X = XYZtoRGB Y G B Z where XYZtoRGB is a 3 by 3 transformation. White is used to scale the values. The resulting transformation relates the monitors RGB values and the standard XYZ coordinates. The NTSC provides a standard set of (x,y,z) chromaticities: red = (0.67, 0.33, 0.0), green = (0.21, 0.71, 0.08), blue = (0.14,0.08,0.78) and white = (0.313, 0.329, 0.358). The resulting transformations from XYZ to RGB are, R 1.967 0.548 0.297 X G = 0.955 1.938 0.027 Y B 0.064 0.130 0.982 Z and from RGB to XYZ: X 0.589 0.179 0.183 R = 0.290 0.605 0.104 G Y Z 0 0.068 1.020 B

3.5.5 HSI Color Space

To produce several shades of blue ranging from light blue (almost white) to dark blue (almost black) is difficult using the RGB, CMY, YIQ or XYZ device-based color spaces. Perceptually-based colors spaces were developed to provide a more intuitive way to define colors for graphics programming. Perceptually-based color spaces use hue, saturation, and intensity (or brightness, lightness, value) as color coordinates. Hue is the basic component of color and is responsible for the perceived color sensation. It is directly related to the dominant wavelengths of the given color. It is important to keep in mind that wavelengths are a physical property and that color is a psychological effect. All color perceived is a result of the stimulation of the photo-receptors in the visual system. Thus, the presence of reddish and greenish wavelengths of light does not produce yellowish wavelengths light, even though the resulting light would be perceived as yellowish because of the way individual sources stimulate the retina.

Chapter 3. Color

3.5 Color Representations

Saturation and chroma describe the purity of the color, or its colorfulness. The greater the presence of achromatic light (light with a horizontal spectral curve), the less saturated a color becomes. As saturation decreases, the dominant hue becomes more difficult to discern. Chroma is the colorfulness compared to white or the level of illumination. Gray is unsaturated, whereas pure monochromatic red of low intensity has high saturation and low chroma. Saturation generally is independent of intensity, whereas chroma depends on intensity. Lightness and Brightness are often used in describing color. Brightness varies from black to bright, representing the strength of the stimulation as if the stimulant is self-luminous. Lightness, on the other hand, varies from black to white representing the reflected intensity of a diffuse surface that is externally illuminated. Thus, as the intensity of the illuminating source increases, the color of the surface becomes de-saturated as it approaches white in color. Here, we use intensity for lightness. The hue-saturation-intensity (HSI) color space, also called hue-saturation-lightness (HLS), is a popular perceptually-based color space that is a double cone with either flat sides (called a hexcone as shown in Figure 3-16) or as a rounded double inverted cone with curved sides. Colors I 1.0 White

Green (2/6 or 120 ) Cyan (3/6 or 180 ) Blue (4/6 or 240 ) 0.5

Yellow (1/6 or 60 ) Red (0) Magenta (5/6 or 300 )

0.0 Black

H S

Figure 3-16. HSI Color Space.

Chapter 3. Color

3.5 Color Representations

are points within the volume of the double cone. Some representations draw the space as a cylinder with black on the bottom surface and white on the top surface. Think of slicing the color cone at a given intensity, producing a two-dimensional color wheel of hues varying around and saturations varying radially. Hue (H) is the angular coordinate. It is represented by the angle around the equator of the cone, normalized to lie in the range [0,1] or [0,360]. Note that this does not follow physical wavelengths exactly. There is no reddish-blue in the electromagnetic spectrum. Saturation (S) is the radial coordinate. Full (1.0) saturation is what one might describe as a deep color. Colors with zero saturation, shades of gray, are located along the axis connecting the tips of the cone. Full saturation is on the outer surface of the cone. Intensity (I) is the vertical coordinate. Zero intensity of any hue and saturation is black. One-half intensity is along the equator. Full (1.0) intensity of any saturation and hue is white. The pure colors (fully saturated, without achromatic light) lie in the plane at half intensity. A color given in HSI coordinates is converted to RGB coordinates in a two step process. The first step is to compute the relative RGB values, (R,G,B) using only the hue (H) coordinate. This is similar in nature to finding chromaticity ratios without regard to luminance. These values can be represented as periodic functions that cycle through the visible light spectrum as approximated by the RGB primaries. The linear versions of the functions, corresponding to the hexcone, are tabulated below and are graphed in Figure 3-17. H 0 to
1 -6 2 -6 3 -6 4 -6 5 -6 1 -6 1 -6 2 -6 3 -6 4 -6 5 -6

R 1 2-6H 0 0 6H-4 1 2-6H

G 6H 1 1 4-6H 0 0 1

B 0 0 6H-2 1 1 6-6H 0

to to to to

to 1 to
2 -6

RGB are converted to RGB as follows: For 0 <= I <= 0.5 : RGB = 2 I ( 0.5 - S ( RGB - 0.5 ) ) For 0.5 < I <= 1: RGB = 0.5 + S ( RGB - 0.5 ) + ( 2 I - 1 )( 0.5 - S ( RGB - 0.5 ) )

Chapter 3. Color

3.5 Color Representations

1 R 0 0 1 G 0 0 1 B 0 0
1 -6 2 -6 3 -6 4 -6 5 -6 1 -6 2 -6 3 -6 4 -6 5 -6 1 -6 2 -6 3 -6 4 -6 5 -6

H Figure 3-17. R B Ratio Functions. G As an example, consider creating realistic images of objects by simulating of the effects of light reflecting from surfaces. White light reflecting from an object of a certain color produces different shades of that color. The problem, then, is to produce a range of shades of a given color, i.e. different intensities of a basic hue and saturation. HSI is a convenient color space for these computations. Given the hue and saturation of the basic color, vary the intensity (I) between 0 and 1 and produce an RGB for each. For example, the following table shows five shades of the color red. I 1 2 3 4 5 0.00 0.25 0.50 0.75 1.00 R 0.0 0.5 1.0 1.0 1.0 G 0.0 0.0 0.0 0.5 1.0 B 0.0 0.0 0.0 0.5 1.0

Examine the color paths of through the five colors in the HSI and RGB color spaces (Figure

Chapter 3. Color

3.5 Color Representations

3-18). 5
White

3 2 1 Black
B

5 1 2 4 3
R

Figure 3-18. Spectrum Colors in HSI and RGB Spaces.

3.5.6 HSV Color Space

The Hue-Saturation-Value (HSV) space, also called Hue-Saturation-Brightness (HSB), is similar to the HSI color space. Geometrically, there is a single cone and the center of the top circle represents white (Figure 3-19). A color given in HSV coordinates is converted to RGB coordinates V Green (2/6 or 120 ) Cyan (3/6 or 180 ) Blue (4/6 or 240 ) 1.0 White Yellow (1/6 or 60 ) Red (0) Magenta (5/6 or 300 )

0.0 Black

H S

Figure 3-19. HSV Color Space. in a two step process. The first step is to compute the RGB ratios as described previously for the

-x

-x,-y

-y Step Size

+x,-y

Figure 4-2. Possible Pen Steps. There are 8 possible steps corresponding to horizontal, vertical and diagonal pen motions. Support software in the plotting package, i.e. in plot, must create each line as a sequence of these steps. This is accomplished by a stepping algorithm, the most famous of which is the Bresenham algorithm [Bresenham]. Of particular note with this algorithm is that it requires only addition and subtraction of integers, so it can be implemented efficiently and inexpensively in software or hardware with simple processors. The first plotters accepted only simple binary commands that specified step directions and pen control, so line generation had to be done by the host computer software. Now, most plotters do this internally, so that only endpoint coordinates of the line need to be sent by the host computer.

Chapter 4. Basic Graphics Software

4.2 Drawing Lines

4.2.1 Bresenham Line Stepping Algorithm

Consider the situation shown in Figure 4-3, where the pen is at location (0,0) and a line is to be drawn whose exact equation is: u y = v x, where u is x and v is y in steps. Any line that Y jn + 1 Exact line: u y = v x A major step E

minor step n

(0,0)

n+1

Figure 4-3. Stepping Algorithm Variables. does not pass through the origin can be transformed into such a line using a simple change of variables. Assume that the pen has been moved (stepped) to the position labelled P. There are two possible next steps: a major diagonal step to point A and a minor horizontal step to point B. The term major indicates that the step involves both X and Y, whereas a minor step involves only one direction. The exact coordinates of the point on the line at x = n+1 is point E, whose y value is v ( n + 1 ) -- . To decide which step, minor or major, would be closer to the exact line, first compute u two positive vertical distances, AE and EB: v AE = (jn + 1 ) - ( n + 1 ) -u v EB = ( n + 1 ) -- - jn u Let the quantity dn+1 be the scaled difference of these two distances (assuming u > 0):

Chapter 4. Basic Graphics Software

4.2 Drawing Lines

dn+1

= u { EB - AE } v v = u { [ (n+1) -- - jn ] - [ (jn + 1 ) - ( n + 1 ) -- ] } u u = 2v ( n + 1 ) - u ( 2 jn + 1 ).

For the previous step (substituting n-1 for n): dn = 2nv - u ( 2 jn-1 + 1 ).

Now express dn+1 in terms of dn to formulate the incremental stepping algorithm dn+1 and = dn + 2 v - 2 u ( jn - jn-1 )

d1 = 2v - u (when n = jn= 0).

Note that ( jn - jn-1 ) = 0 if dn < 0, which means take a minor step and dn+1 = dn+ 2v, = 1 if dn >= 0, which means take major step and dn+1 = dn+ 2v - 2u For lines in the other four quadrants (and when y > x), it is necessary to define the proper minor and major steps according to the particular plotter and its commands, and the proper u and v quantities using absolute values. For example, given a line with a negative x and positive y, i.e. with the endpoint in the second quadrant, the minor step would be -x and the major step would be -x+y. The quantities u and v must be computed as their first octant values. This algorithm is summarized in Figure 4-4. major = proper plotter command for diagonal step; minor = proper plotter command for minor step; u = max( abs(dx), abs(dy) ); v = min( abs(dx), abs(dy) ); d = 2v - u; for( counter = 0; counter < u; counter++ ) { if( d >= 0 ) { d = d + 2v - 2u; /* or d += 2v - 2u; */ PlotterCommand(major); } else { d = d + 2v; /* or d += 2v; */ PlotterCommand(minor); } } Figure 4-4. Stepping Algorithm Logic.

Chapter 4. Basic Graphics Software

4.2 Drawing Lines

Note that the quantities 2v - 2u and 2v are constants, and 2v = v+v. Figure 4-5 is an example of the execution of the algorithm. DX = 5, DY = 4, therefore: u = 5, v = 4, major = +x+y , minor = +x exact line dn dn+1 step 4 3 1 -1 7 5 1 -1 7 5 3 major 3 2 3 4 5 major minor major 1 major 0 0 1 2 3 4 5 2 actual steps

n 1

Figure 4-5. Example Steps. Figure 4-6 is an implementation of the stepping algorithm for a plotter whose step command is an integer in which each bit has the meaning shown in the figure. PlotterCommand is an operating system interface routine that sends the given integer, assumed to be a plotter command, to the plotter.

4.2.2 The Plot Routine

As an example of the internal operation of a plotting package, examine how the basic plot routine could be coded. To begin, it seems logical that a user would not want to use coordinates in units of plotter steps. This would be very inconvenient and makes the program specific to a particular plotter. Also, the user may want to easily change the plot scale (for example: to reduce the picture size for publication) and origin (for example: to produce a series of side by side graphs). Building on the stepping routine, Step, subroutine Plot could be written as shown in Figure 4-7. The variables xpen, ypen, xorg, yorg, factor, xres, yres are global variables maintained by the plotting package. Xpen and ypen are the current pen location of the plotter. Xorg and yorg store the origin and factor is the scale factor. Xres and yres are the resolutions, which can be varied by the package as different plotters are selected

Chapter 4. Basic Graphics Software

4.2 Drawing Lines

void Step( int dx, int dy ) { int major, minor, u, v, d, count;

plotter command: u = dx; if( u < 0 ) { 32 16 8 4 2 1 u = -u; // abs(dx) major = 4; // -x PEN PEN +X -X +Y -Y DN UP } else major = 8; // +x example: 0010012 = 910 = step +X -Y v = dy; if( v < 0 ) { v = -v; // abs(dy) major |= 1; // OR -y bit } else major |= 2; // OR +y bit if( u >= v ) { // major direction is x minor = major & 12; // save only x bits } else { // else major direction is y minor = major & 3; // save only y bits d = u; // and swap u & v u = v; v = d; } count = u; v += v; // 2v for minor d increment d = v - u; // initial d is 2v - u u = d - u; // 2v - 2u for major d increment while( --count >= 0 ) { if( d >= 0 ) { d += u; PlotterCommand( major ); } else { d += v; PlotterCommand( minor ); } }
}

Figure 4-6. An Implementation of the Stepping Algorithm. for use. The routines EndPlot, PenUp and PenDown are system interface routines that communicate with the plotter (on-line) or enter plotter control commands to a plot file (off-line).

Chapter 4. Basic Graphics Software

4.3 Characters

/* global package data */ float xpen, ypen, xorg, yorg, factor, xres, yres; void Plot( float x, float y, int pen ) { int dx, dy, penabs; float xold, yold; if( pen == ENDPLOT ) { Endplot(); return; } xold = xpen; yold = ypen; xpen = x * factor + xorg; ypen = y * factor + yorg; penabs = pen; if( pen < 0 ) { penabs = -pen; xorg = xpen; yorg = ypen; } if( penabs == MOVE ) Penup(); if( penabs == DRAW ) Pendown(); dx = ceil( xres * (xpen - xold) ); dy = ceil( yres * (ypen - yold) ); Step(dx, dy); }

Figure 4-7. An Implementation of Plot.

4.3 Characters
To draw a character requires that its graphic appearance, or glyph, is pre-defined in some drawable form. It is necessary to create a general definition of each character that can be scaled and positioned as needed. We establish a definition grid, a layout convention for defining characters. A complete set of character definitions is called a font.

4.3.1 Character Definitions

Look at an example, assuming a 7 by 7 definition grid to define a capital 'A' as shown in Figure 4-8. The routine to plot the symbol A is coded in Figure 4-9. The variable f is a scale factor that scales grid-units to inches.

Chapter 4. Basic Graphics Software

Data Arrays: xgrid[] ygrid[] pen[]

data for A

max. #char

data for B

max. #points for ALL chars. Figure 4-11. A Font Data Structure. 1. ASCII order, fixed size. The character definitions in the data arrays must appear in increasing numerical order according to the ASCII codes. Also, the definitions must contain the same number of points, which we will assume is given in a predefined constant, CHARLEN. Therefore, the dimensions of xgrid, ygrid and pen are 128*CHARLEN (assuming 128 characters, with ASCII codes 0 to 127, will be stored). The data for a given character can be found using only the ASCII code. The index of the start of the definition for the character with ASCII code ascii is computed as CHARLEN * (ascii - 1), and the definition has CHARLEN points.

ASCII order, variable size. The definitions appear in increasing ASCII order, but the sizes of the definitions are

allowed to vary. The is an improvement over the fixed size approach because now each character can be designed without concern for exceeding the size limit or wasting valuable data space. However, the starting index in the data arrays can no longer be computed from the ASCII code alone. Another data array must be created to store the starting indices in the data arrays: the start table. This table is indexed by the ASCII code. Due to the fact that the definitions are in ASCII order, the end of one definition can be computed from the start of the definition for the character with the next ASCII code.

Chapter 4. Basic Graphics Software

4.4 Raster Software

Unordered, variable size. Requiring ASCII order can be overly constraining. The data can

be more easily created and edited, for example, if definitions are allowed to appear in any order. The definitions are no longer in ASCII order, so the end of a definition can no longer be computed as above. This makes it necessary to extend the table with another data array to store the end index of each definition, the end array. (Some may find it more convenient to store the size of each character.) The arrays start and end are indexed by the ASCII code. Figure 4-12 shows how routine Symbol could be coded using these data structures.
void Symbol( float x, float y, float ht, char *string, int ns ) { float f, dx; int ascii, i1, i2, i, j; static int xgrid[?]={...}, ygrid[?]={...}, pen=[?]{...}; static int start[?]={...}, end[?]={...}; f = ht / 7.0; dx = 0.0; for( i = 0; i < ns; ++i ) { ascii = string[i]; /* option 1: ascii order, fixed size */ i1 = CHARLEN*(ascii-1)+1; i2 = i1 + CHARLEN - 1; /* option 2: asci order, variable size */ i1 = start(ascii); i2 = start(ascii+1)-1; /* option 3: unordered, variable size */ i1 = start(ascii); i2 = end(ascii); for( j = i1; j <= i2; ++j ) plot( x+dx + f*xgrid[j], y+f*ygrid[j], pen[j] ); dx = dx + ( ht * 6.0 / 7.0 ); } }

Figure 4-12. Symbol Using A Font Data Structure.

Chapter 4. Basic Graphics Software

4.4 Raster Software

discussed previously, but now a step means changing running pixel coordinates and setting the pixel value. For example, a horizontal minor step means incrementing or decrementing the current h coordinate along the line, and setting the pixel value at this (h,v) to the desired pixel value. Thus, a line is drawn by operating on the computed set of pixels that best approximates the theoretical line between two pixels. Note that for a raster display, there is no concept of a line in pixel memory. The display data representing the image are only the pixel values in pixel memory that are being refreshed on the CRT screen. Drawing a line means rasterizing the concept of a line into the appropriate set of pixels in pixel memory. Only the software doing the rasterization understands that there is a line. In addition to lines, there are other drawing operations that can be performed on raster displays to take advantage of their dynamic nature. One common raster operation is filling an area, like a rectangle, with a given pixel value. The location and dimensions of a rectangle are specified in different ways, for example as the upper left (h,v) coordinates and the width and height (as in X windows), or alternatively the upper left and lower right coordinates (as in Macintosh Quickdraw). Given the rectangle data, we simply use nested loops to set every pixel within the rectangle boundaries. An interesting variation of the typical drawing operations is drawing pixel patterns or textures. Typically, a pattern is an ordered sequence of varying pixel values specified over an area. For example, 50% gray would be alternating black and white pixels across and down scan-lines. Patterns could be pictures or other sequences of rows and columns of pixel values that can be painted into an area. It is possible in most packages to specify a pattern for the pen that is applied to lines or filled areas. Now consider more general operations that just setting a pixel value to draw an object. In general, raster drawing involves computing the pixel locations for a given drawing operation. For these pixel locations, a given pixel value, called the source pixel, is copied into pixel memory at each computed pixel location, called the destination pixel. For example, to draw a line, destination pixel locations are computed using the Bresenham stepping algorithm, and the source pixel value, often called the current pen color by graphics packages, is copied into each of these locations. Filling is similar, but the destination pixels are more easily computed. In principal, we could perform any arithmetic or logical operation on the source and/or destination pixel value, in addition

Chapter 4. Basic Graphics Software

4.4 Raster Software

to the one we have used to this point, copy. In fact, raster graphics packages typically provide a number of pixel transfer functions, as they are called. One that is particularly useful for interaction is the exclusive-or, or XOR. The XOR operation is a well known Boolean logic function whose binary logic table is shown in Figure 4-15. In short, XOR is binary addition without carry. An important property of the A value 0 1 0 1 1 0

B value

0 1

Results of A XOR B

Figure 4-15. XOR Truth Table. XOR operator is that XORing the same source value twice to a destination value restores the original destination value. That is, for two numbers A and B, A XOR (A XOR B) = B. A very useful application of XOR for dynamic graphics is to draw and erase an object by drawing it twice at the same location using the XOR transfer function. Doing this repeatedly while moving the object creates the effect of moving the object over other objects on the screen, called dragging, without the need to redraw the other objects. Notice that for the pixel values of black and white, this works neatly as illustrated in Figure 4-16. For white = 0 and black = 1: source black(1) black(1) destination white(0) black(1) source xor destination black(1) white(0)

Figure 4-16. XOR Operations on a Black and White Display. For color displays, where pixel values are essentially arbitrary integers, the XOR operation will produce arbitrary colors. Consider the case of a mapped display, where pixel values are color table indices. Suppose that the color white is the pixel value 5 and black is 6. Drawing a black line over a white area means the source pixel values are black, or 6, and the destination pixel values are white, or 5. Drawing the line in XOR mode means that the system XORs each source and destination pixel value for each pixel location along the line. In this case, 5 XOR 6 is 3 (5 is 1012,

Chapter 4. Basic Graphics Software

4.5 References

6 is 1102, 1012 XOR 1102 = 0112 or 3 ), so the resulting pixel values along the line are 3. But what color is pixel value 3? For a mapped display, it is the color in color table location 3, which may or may not be a color set by you! Drawing the same line in black a second time means that the pixel values will be 6 XOR 3, or 5, which restores the color white, as expected. XOR is useful only during temporary graphic operations where efficiency is essential, such as dragging. After XOR operations are finished, such as completing the dragging of an object, the entire image must be re-drawn with the copy transfer function to give the correct image.

4.5 References
1. J. E. Bresenham, Algorithm For Computer Control of A Digital Plotter, IBM Systems Journal, Volume 4, No. 1, 1965.

Chapter 4. Basic Graphics Software

4.6 Review Questions

1. The Bresenham algorithm computes steps that approximate a straight line. The algorithm can be adapted for different devices.What is the Bresenham step for a pen plotter? What is the Bresenham step for a raster display? 2. Explain in detail how the GRAFIC pen mode GXOR allows objects to be dragged across the screen without erasing the underlying picture. Give before and after examples with pixel values for a black and white display, and a mapped color display. 3. Some raster displays have pixel values of black=1 and white=0, while others have white=1 and black=0. If the background color of a given display is white and the foreground (pen) color is black, show the pixel values and explain the visible results of drawing one line in XOR pen mode for each of these cases. 4. How could the font data and code for Symbol in Figure 4-12 be modified to produce proportional spacing? 5. Character fonts for raster displays are often defined as closed boundaries consisting of straight lines. A character instance is drawn by setting all pixels in the interior of its transformed boundary. (a) (b) Diagram and explain the data structure necessary to store such a font. Explain, in step by step form, how to implement the routine

DrawCharacter(ichar,h,v), which draws the character whose ASCII code is ichar at the location (h,v) on the screen using GRAFIC. 6. A low-cost laser printer must be sent an entire 8 by 10 inch page of pixels (in scan-line order). The page is printed after all pixels have been received. Each pixel is represented by a character: 'b' for black, 'w' for white. The printer has x and y resolutions of 300 pixels/ inch. Describe the software needed to produce a printed page on this device, starting with an application program that calls the following routines: 1. STARTPICTURE is called to initialize plotting. 2. LINE( x1, y1, x2, y2 ) draws a straight line from (x1,y1) to (x2,y2), where the coordinates are given in (real) inches from the lower left of the page. 3. ENDPICTURE is called to signify plotting is finished.

Chapter 4. Basic Graphics Software

4.6 Review Questions

(A) Draw a block diagram showing the general organization of the software, including the stages of processing, types of data, and flow of information starting with the application program and ending with the printing process. (B) Describe in detail the process of creating data for the printer after the application program has completed.

Chapter 4. Basic Graphics Software

5.1 Display Programming

Chapter 5. Vector Display Systems

We have seen that programming for a raster display processor is a matter of setting pixels in a frame buffer. Now consider the vector display processor, a point plotting CRT with a vector generator. The vector generator is special hardware that controls the beam intensity and location to draw a line as a continuous stroke from one screen location to another. The vector display processing unit, or vector DPU, connects to a computer and responds to commands analogous to an on-line pen plotter. The DPU commands, called display instructions, are for drawing lines and characters, controlling refresh and communicating with a host computer. Creating images for a dynamic vector DPU is a very different matter than for a static device like a plotter. To have the DPU draw a picture one must program it. This is called display programming.

5.1 Display Programming

Assume we have taken delivery of a new vector display system that is driven from a host computer by instructions that are sent over a high-speed interface. The DPU, then, simply responds to instructions it receives from the host, one at a time.

Host Computer (Program)

Vector Display Processor

Display

Figure 5-1. Vector Display Configuration. We study its programming manual and learn that the screen coordinates range from -2048 to +2047 in x and y, with [0,0] at the screen center. The instructions must be sent as binary data to the DPU, but for convenience to humans they have been given symbolic forms, or mnemonics. MOVEA x,y MOVER dx,dy DRAWA x,y DRAWR dx,dy move to the absolute 2D point (x,y) move by dx and dy draw to the absolute point (x,y) draw by dx and dy

Chapter 5. Vector Display Systems

5.1 Display Programming

data. We modify the support software in the host to perform this new data communication: ToDisp ( int array[], int nwords, int addr ); send nwords of display instructions to the display, loading them from array and storing them starting at display address daddr. The beam movement for this program is shown in Figure 5-9. main() { static int dlist[10] = { /* list of integers corresponding to the */ /* display instructions in Figure 5-7 */ }; ToDisp( dlist, 10, 1); } Figure 5-8. Two Squares Display Driver. Y

500,500

100,100 X Figure 5-9. Beam Movement for Two Squares Display Program.

Chapter 5. Vector Display Systems

5.2 Dynamic Display Manipulation

What happens when the program terminates? The boxes remain displayed because the display is performing the refresh of the display instructions from its own memory.

5.2 Dynamic Display Manipulation

Expand this driving program to interact with a device, in this case an analog-to-digital converter that is interfaced to the same computer as the display.

Host Computer (program)

DPU

CRT

Analog-Digital Converter

Channel 1 Channel 2 "X" "Y"

Figure 5-10. Hardware for Dynamic Display Manipulation. Our interface routines, developed as the other, are: int ReadVolt( int ichan ); returns a numerical value that is the current voltage read from input channel ichan. int MakeMove( int x, int y ) ; returns a MOVEA x,y display instruction given an x and a y coordinate. The new driving program is shown in Figure 5-11. What happens when the potentiometers are moved? The first box moves, because the program over-writes the MOVEA instruction at location 1 in display memory with a new instruction that was made by the MakeMove function from the x and y values read from the potentiometers.

Chapter 5. Vector Display Systems

5.3 Display Processors

main() { static int dlist[10] = { /* data as before */ }; int x, y; ToDisp( dlist, 10, 0 ); for(;;) { x = ReadVolt( 1 ); y = ReadVolt( 2 ); /* Note: C arrays start at index 0 */ dlist[0] = MakeMove( x, y ); ToDisp( dlist, 1, 1 ); } } Figure 5-11. Display Driver for Dynamic Display Manipulation. Note that the display list is being changed as the DPU executes it. This is dynamic display manipulation. One can now think of many other desirable functions: changing the size of the box, adding new boxes, deleting boxes pointed to by a cursor that we move with the potentiometers, etc. This involves the development of a graphics package of routines which provide a variety of functions: 1. 2. 3. manage display memory, provide drawing functions for creating entities like lines, characters, etc. provide logical organization of the display to permit adding, deleting, and other operations on groups of entities, 4. provide input functions for obtaining screen positions from devices.

5.3 Display Processors

There are many different types of raster and vector display processors. We will examine two generic configurations. First, introduce the elements of display systems: CPU CPU memory DPU DPU memory central processing unit (hardware used to compute), the program & the data, display processing unit (hardware used to display), the display list,

Chapter 5. Vector Display Systems

5.3 Display Processors

refresh update

data path where the DPU executes display list, data path where the host computer changes the display list.

Case 1: Remote Display Processor with Local Memory This system was discussed earlier during the driving program examples.

CPU

DPU

interface CPU memory update DPU memory refresh

CRT

Figure 5-12. Remote DPU with Memory. Generally, this display processor can be a raster DPU or a random vector DPU. As a Random Vector DPU: DPU memory consists of structured display commands, a display list. As a Raster DPU: DPU memory is unstructured pixel memory. In some cases, in addition to controlling the beam to display pixels, the DPU can also act as a general purpose computer, accepting structured commands from the host, such as line and character . These commands are simply translated into operations on pixel memory. Their meaning is lost in DPU memory. Case 2: The Raster Refresh Buffer This is an interesting variation on a raster display that incorporates a computer into the display to simulate structured display operations. The DPU display list is maintained by DPU #1, an internal computer, which interprets structured instructions from the host, stores them in DPU memory #1 as a structured display list, and translates them into pixels in DPU memory #2, pixel memory. DPU #2 is a simple raster DPU that refreshes the screen from pixel memory. The translation process is termed repaint. The host computer can draw lines and then erase them. This causes DPU #1 to edit its display list, and then repaint pixel memory accordingly. The advantage of such a system is that a raster display now has structured commands, such

Chapter 5. Vector Display Systems

5.3 Display Processors

CPU

DPU #1 (vector) interface

DPU #2 (raster)

CRT DPU memory #1 (display list) repaint DPU memory #2 (pixel memory) refresh

CPU memory

update

Figure 5-13. Raster Refresh Buffer. as lines and characters, and erasure. The display independently handles the repaint process that all raster drawing requires. The disadvantage is the cost and complexity of the additional hardware. Also, in the limit as the update rate approaches the refresh rate, such as in fast animation, one is limited by the speed of DPU #1.

Chapter 5. Vector Display Systems

6.1 Window to Viewport Mapping

Chapter 6. 2D Graphics Software

Now we extend the basic graphics capabilities by developing some fundamental mathematical algorithms and methods for 2D graphics. These methods add to the previous basic graphics capabilities to provide more sophisticated functionality.

6.1 Window to Viewport Mapping

To begin, consider the task of drawing data read from a file into a known plotting area. The problem is that the program does not know the data coordinates until the data is read. The user, i.e. the creator of the data, could be forced to manipulate (transform) the data to fit in the plotting area, but this is generally unacceptable and far too device specific. There is a better, more general, approach based on the graphical method known as window to viewport mapping (Figure 6-1). First, establish some terms: window: viewport: a rectangle defined in data units, or world coordinates. a rectangle defined in device units, or device coordinates. window y v yw xv xw Figure 6-1. Window and Viewport. viewport image

6.1.1 Mapping Equations

The problem is to map, or transform, world coordinates into viewport coordinates so they can be drawn on the device. The variables that represent the location and dimensions of each space are shown in Figure 6-2. Think of mapping the window boundary to the viewport boundary. The derivation of the mapping from coordinates in window units, (xw,yw), to the corresponding

Chapter 6. 2D Graphics Software

6.1 Window to Viewport Mapping

window wsx

viewport Yv Xv wsy

(xw,yw)

(wcx,wcy)

(xv,yv)

vsx vsy

(vcx,vcy)

Figure 6-2. Window and Viewport Variables. coordinates in viewport units, (xv,yv), using the window and viewport parameters involves three steps: 1. Translate to window coordinates relative to (wcx, wcy): x1 = xw - wcx y1 = yw - wcy 2. Scale from window to viewport coordinates relative to (vcx, vcy): vsx x2 = x1 -------wsx vsy y2 = y1 -------wsy 3. Translate to viewport coordinates relative to the viewport origin: xv = x2 + vcx yv = y2 + vcy Combine these into single equations: vsx xv = (xw - wcx) -------- + vcx wsx vsy yv = (yw - wcy) -------- + vcy wsy As a check, examine the units of the last equation for xv and yv: ( device units ) (device units) = (data units - data units) ---------------------------------- + (device units) ( data units ) As expected, we get (device units) = (device units).

Chapter 6. 2D Graphics Software

6.1 Window to Viewport Mapping

As a further validation select a test point, the upper right corner of the window, and map it to viewport coordinates. The coordinates of the upper right corner are: (wcx+wsx, wcy+wsy). Substituting these for xw and yw into the equations above, vsx xv = (wcx+wsx - wcx) -------- + vcx wsx vsy yv = (wcy+wsy - wcy) -------- + vcy wsy These are the expected values. = vsx + vcx = vsy + vcy

6.1.2 Uniform Scaling

vsx vsy If the two scale factors -------- and -------- are not equal, i.e. the aspect ratios (width/height) of wsx wsy the window and viewport are not equal, the data is unequally scaled in x and y, causing distortion (Figure 6-3). window viewport

Figure 6-3. Distortion Due to Unequal Scaling in X and Y. This can be avoided by altering the mapping process to perform uniform scaling. This vsx vsy involves using a single scale factor, factor = min( -------- , -------- ) , in place of two separate factors in wsx wsy the second step. In this case, the window boundary will map on or inside the viewport boundary. With uniform scaling, the window may map into a rectangle smaller than the viewport, depending upon the aspect ratios of the two. Using the center of the window and viewport as the reference point in the equations above causes the data to remain centered in the viewport. In many applications, the window is computed from the given data to be the smallest enclosing rectangle, often called the extent. The extent is computed by scanning the data coordinates to find the extrema for x and y, which are the coordinates of the diagonals of the

Chapter 6. 2D Graphics Software

6.2 Clipping

bounding rectangle: (xmin, ymin) and (xmax, ymax). From these coordinates, the appropriate window variables are easily found. extent rectangle (xmax,ymax)

data

(xmin,ymin) Figure 6-4. Data Extent.

6.2 Clipping
Window to viewport mapping transforms coordinates from one coordinate system to another. Coordinates outside the window map to coordinates outside the viewport. It is often desirable or necessary to remove or clip portions of the data that lie outside a given boundary. At first glance, one might be tempted to write a routine with a long and complicated sequence of if tests, each dealing with a special case where the line is outside the window. Fortunately, there is a more organized approach. An elegant algorithm, known as the Cohen-Sutherland clipping algorithm, was developed for line clipping. The algorithm is based on a special integer code, called a region code or endpointendpoint code, that is computed for each endpoint of a line. This code compactly designates the boundary of the 2D region or zone in which the point lies, and has been designed for ease of use and efficiency in handling special cases.

6.2.1 Region Codes

Two-dimensional space is divided by four boundaries into 9 regions, all of which are semiinfinite (extend in one direction to infinity) except one, which is designated the visible region. Each boundary is assigned a bit code, or power of two integer, uniquely designating the violated boundary. The visible region is the center region and has a code of 0. One possible coding scheme

Chapter 6. 2D Graphics Software

6.2 Clipping

is shown below: TOP 10002 (810) BOTTOM 01002 (410) RIGHT 00102 (210) LEFT 00012 (110) IN 0

The region codes are computed based on these conditions as shown in Figure 6-5.

10012

10002

10102 ymax

00012 visible region 01012 xmin

00002

00102 ymin

01002

01102 xmax

Figure 6-5. Clipping Region Codes. Code to compute the clipping region code is illustrated in Figure 6-6. Note that points on the code = IN; if( x < xmin ) code = LEFT; else if( x > xmax ) code = RIGHT; if( y < ymin ) code = code + BOTTOM; else if( y > ymax ) code = code + TOP; Figure 6-6. Clipping Region Computations. boundary of the visible region, as well as those inside, have a region code of IN. Instead of attempting to detect and process the many possible cases of lines with endpoints in different regions, the algorithm uses the region codes to determine if the line is entirely visible or entirely invisible. If not, then one of the endpoints that is outside any boundary is re-computed to be the intersection of the line and the boundary, sometimes called pushing the endpoint to the boundary. This clipped line is then subjected to the same process. This test and clip iteration continues until the re-computed line is entirely visible or invisible. If the resulting line is visible, it

Chapter 6. 2D Graphics Software

6.2 Clipping

is drawn. If it is invisible, it is ignored. The region codes greatly simplify and speed the visibility tests on a line. A line is visible when both endpoint codes are zero, a simple and fast computation. The value of the region bitcoding scheme is seen in the test for an entirely invisible line. Without the region codes, one could code this test as an unwieldy statement of the form if both endpoints are above the top boundary of the visible region, or if both endpoints are below the bottom boundary of the visible region, if both endpoints are left of the left boundary of the visible region, or if both endpoints are right of the right boundary of the visible region, then the line is entirely invisible. With the region codes, however, this test is simple. Notice that the condition of both endpoints violating a boundary, i.e. above the top, below the bottom, right of right, left of left, means that the region codes will have 1s in corresponding bit positions. This means that the integer and of the two region codes will be non-zero. In C, the "and" computation is the & operator, result = i1 & i2; The invisibility test reduces to a simple computation and test against zero. In effect, the and operation is four parallel if statements.

6.2.2 Clipping Against A Boundary

Computing the clipped endpoint when a line crosses a clipping boundary is a geometric computation that is best done using the symmetric equation of a line. Given a line with endpoints (x1,y1) and (x2,y2), any point (x,y) along that line satisfies the following equation, y y1 x x1 ---------------- = ---------------- . x 2 x1 y 2 y1 For a given boundary, either x (x = xmin for the LEFT boundary and x = xmax for the RIGHT) or y (y = ymax for TOP and y = ymin for BOTTOM) will be known, so the equation can be solved for the other coordinate. There are four boundaries, so there will be four pairs of equations. For example, if endpoint 1 of line crosses the RIGHT boundary, then endpoint 1 would

Chapter 6. 2D Graphics Software

6.2 Clipping

be re-computed (clipped) as follows, y2 y1 y 1 = ---------------- ( xmax x 1 ) + y 1 x2 x1 x 1 = xmax Note that the order is critical. The y1 computation must use the original x1 coordinate. Although the algorithm appears at first to clip unnecessarily many times, for example, a line may be clipped several times before being found to be outside the visible region, in fact it is efficient and very compact. Figure 6-7 illustrates how a line is processed during the iterations of the algorithm. 10 Pass Step 1 0 1 2 11, 22, 23 13 Step 3 12 Step 2 20 , 21 Figure 6-7. Illustration of the Iterations of the Clipping Algorithm. 3 Endpoints 10 - 20 11 - 21 12 - 22 13 - 23 Action Original Line 10 was LEFT 21 was RIGHT 12 was BELOW

The resulting line 1323 is visible.

Chapter 6. 2D Graphics Software

6.3 2D Transformations

An implementation of the algorithm is shown in Figure 6-8.

void ClipLine2D( float x1, float y1, float x2, float y2 ) { int c1, c2, ctemp; float ftemp; c2 = RegionCode( x2, y2 ); for(;;){ c1 = RegionCode( x1, y1 ); if( c1 == IN && c2 == IN ) { /* process visible line from (x1,y1) to (x2,y2) */ return; } if( (c1 & c2) != 0 ) { /* the line is invisible - skip or notify caller*/ return; } if( c1 == IN ) { /* swap points 1 and 2 to simplify cases */ ctemp = c1; c1 = c2;c2 = ctemp; ftemp = x1; x1 = x2;x2 = ftemp; ftemp = y1; y1 = y2;y2 = ftemp; } if( c1 & LEFT ) { y1 += ( y2 - y1 ) / ( x2 - x1 ) * ( xmin - x1 ); x1 = xmin; } else if( c1 & RIGHT ) { y1 += ( y2 - y1 ) / ( x2 - x1 ) * ( xmax - x1 ); x1 = xmax; } else if( c1 & TOP ) { x1 += ( x2 - x1 ) / ( y2 - y1 ) * ( ymax - y1 ); y1 = ymax; } else { /* BOTTOM */ x1 += ( x2 - x1 ) / ( y2 - y1 ) * ( ymin - y1 ); y1 = ymin; } } }

Figure 6-8. Implementation of the Clipping Algorithm.

6.3 2D Transformations
Window to viewport mapping involved coordinate transformations. In general, a transformation is a mapping: 1. of one set of points into another in a fixed reference frame, (Figure 6-9), i.e object

transformations, 2. from one coordinate system to another (Figure 6-10). This is termed

change of basis. For now, focus on the object transformations, the mapping of a set of points (data) into another.

Chapter 6. 2D Graphics Software

6.3 2D Transformations

Y (x,y)

(x,y)

X Figure 6-9. Object Transformation. Y1 (x1,y1) Y2 (x2,y2) X1

X2 Figure 6-10. Change of Basis Transformation.

6.3.1 Matrix Representation

Algebraic representations for transformations, i.e. variable substitutions in equations, limit computer implementations in several ways: 1. 2. a special procedure is needed for each basic transformation, it is difficult (sometimes impossible) to provide general transformation capabilities in algebraic form in the computer, 3. it is not generally possible to find the inverse of the algebraic equations resulting from a series of transformations. (For example, this is needed to restore the original position of an object after several transformations.) We can represent the equations in a more general and computationally convenient way using matrices. In general, two equations of the form x = a x + b y y = d x + e y are expressed in the matrix equation: a d x' y' = x y b e

Chapter 6. 2D Graphics Software

6.3 2D Transformations

when the coordinates x and y are represented as row vectors. When the coordinates are represented as column vectors, the matrix equation appears as, x' = a b x y' cd y Define symbols for the vectors and matrix: V' = x' y' V = x y M = a d b e

Using row vectors, the symbolic equation is : V = V M, and using column vectors, it is V = MT V, where MT is the matrix transpose of M. Given the symbolic equation, it is generally apparent whether row or column vectors are in use. However, given only a matrix, it is necessary to know if this matrix is to be applied to row vectors (vector before matrix) or column vectors (vector after matrix). These notes will use row vectors for coordinates. For 2D operations, we are tempted to make M 2 by 2, but this disallows constants because all M elements multiply x or y. Expand the representations of M and V. V = x y 1 a d M = b e c f

This will work, but looking ahead we see that having M 3 by 2 will cause problems because we cannot invert a 3 by 2 matrix and we cannot multiply a 3 by 2 by a 3 by 2. For these reasons, we add a third column to M: a d 0 M = b e 0 c f 1 and 1 0 0 I = 0 1 0 0 0 1

6.3.2 Basic Transformations

There are three basic transformations that can be combined to form more complex transformations: translation, scaling, and rotation.

Chapter 6. 2D Graphics Software

6.3 2D Transformations

Translation: Y (x,y) X Figure 6-11. Translation. x = x + Tx y = y + Ty X Y (x,y)

The translation transformation in functional form and matrix form: T ( Tx,Ty ) = 1 0 0 0 1 0 Tx Ty 1

Scaling: Y (x,y) Y (x,y) X Figure 6-12. Scaling. x = Sx x y = Sy y X

The scaling transformation in functional form and matrix form: Sx 0 0 S ( Sx,Sy ) = 0 Sy 0 0 0 1 Note that all coordinates are scaled about the origin. This means that what we perceive as an objects position will change by scaling. Mirror images can be done (for some geomtric forms) using negative scale factors. Distortion, or scaling with different factors in x and y, can be avoided simply by using a single scale factor Sx = Sy = S.

Chapter 6. 2D Graphics Software

6.3 2D Transformations

Rotation: Y y (x,y)

y x x (x,y) X

Figure 6-13. Rotation. Given (x,y) and , rotate by the angle . x = R cos( + ) y = R sin( + ) Expanding, x = R ( cos cos - sin sin ) y = R ( sin cos + cos sin ) and then recognizing that x = R cos and y = sin, x = x cos - y sin y = x sin + y cos The rotation transformation in functional form and matrix form: cos sin 0 R ( ) = sin cos 0 0 0 1 Like scaling, the rotation transformation is about the origin (Figure 6-14) so that an objects position will change. Y (x,y) X X Y (x,y)

Figure 6-14. Rotation Occurs About the Origin.

Chapter 6. 2D Graphics Software

6.3 2D Transformations

For simplicity in writing and reading matrix equations, we generally write matrix transformations using the functional forms that were shown above: T(Tx,Ty) S(Sx,Sy) R() for translation, for scaling, for rotation,

The identity transformations, or null transformations, are those that produce the identity matrix: T( 0,0 ), S( 1,1 ), R( 0 ).

6.3.3 Concatenation (Composition)

A series of transformations can be combined (concatenated) into one. For example, consider scaling an object about a given point. We want to reduce the size of the box without changing its position (a,b) (Figure 6-15). Y (x,y)

(a,b) X Figure 6-15. Scaling About a Point. First carry out the derivation algebraically: 1. Translate so that position (a,b) becomes the temporary origin. (a,b) could be any reference point: x1 = x - a 2. y1 = y - b

Scale to the correct size: x2 = Sx x1 y2 = Sy y1

Translate again to restore the position (a,b): x3 = x2 + a y3 = y2 + b

Notice that x3 is a function of x2 and y2 is a function of x1. Substituting: x3 = (x - a) Sx + a The order is very important. y3 = (y - b) Sy + b

Chapter 6. 2D Graphics Software

6.3 2D Transformations

Any number of basic transformations can be concatenated into one pair of equations, eliminating the intermediate results that would have required extra data storage and computation. Concatenation is easy in matrix form. For the scaling example: 1. 2. 3. translate: T(-a,-b) scale: S(Sx,Sy) translate: T(a,b)

Combining these into a matrix equation: v3 = v2 T(a,b) = { v1 S(Sx,Sy) } T(a,b) = { v T(-a,-b) } S(Sx,Sy) T(a,b) This last equation is easily converted to matrix form by substituting the appropriate matrices. We can define the composite transformation matrix, M as: M = T(-a,-b) S(Sx,Sy) T(a,b) and therefore, v3 = v M. Note again that the order of the matrices is critical! For practice, find M using the symbolic matrices:

M =

1 0 0 Sx 0 0 1 0 0 0 1 0 0 Sy 0 0 1 0 a b 1 0 0 1 a b 1

Sx 0 0 1 0 0 = 0 Sy 0 0 1 0 aSx bSy 1 a b 1 Sx 0 0 0 Sy 0 ( aSx + a ) ( bSy + b ) 1

Multiplying out the matrix equation symbolically yields the two expected algebraic equations: x = x Sx + (-a Sx + a ) = ( x - a ) Sx + a

Chapter 6. 2D Graphics Software

6.4 Review Questions

y = y Sy + (-b Sy + b )

( y - b ) Sy + b

Thus, a single 3 by 3 matrix can represent any combination of basic 2D transformations in a compact, codable and simple form. In the computer, these are all numbers. Functional form provides a convenient way of understanding and even solving transformation equations without doing matrix multiplication. For example, given the transformation equation in functional form, one can visualize the resulting picture by performing each basic transformation by hand on the objects. Try this with the previous scaling about a point example. This is easier (and much faster on a test) than performing the arithmetic.

6.3.4 Inverse Transformations

Suppose one wanted to find the inverse of a series of transformations. Instead of computing the inverse of the concatenated 3 by 3 matrix, utilize the fact that the inverse of a product of matrices (transformations) is the product of the inverses in reverse order. The inverses of the basic transformations can be computed by inspection: T(a,b)-1 = T(-a,-b) S(a,b)-1 = S(1/a,1/b) R()-1 = R(-).

6.4 Review Questions

1. Derive the window to viewport mapping for viewports defined on devices that have the origin at the upper left of the screen, with the horizontal coordinate h increasing right, and the vertical coordinate v increasing down. 2. Derive a two-step window to viewport mapping procedure that first maps world data in the window into a unit viewport with center (0,0) and x and y from -1 to 1 called normalized device coordinates (NDC), and then maps this data to the viewport. This procedure isolates device details from the basic mapping process and is useful when data must be clipped. The data is clipped in NDC, which are independent of both the window and viewport, thus isolating the clipping process from them.

Chapter 6. 2D Graphics Software

6.4 Review Questions

Often graphics packages map world data into "clipping coordinates" prior to performing 2D clipping in order to simplify the computations. The clipping coordinates of visible points lie between 0.0 and 1.0. After clipping, the endpoints of visible portions of lines are mapped to device coordinates and drawn. The process is: (1) apply the transformation [T1] to map world coordinates into clipping coordinates, (2) 2D clip, and (3) apply the transformation [T2] to map the clipped coordinates into device coordinates. Given a square window in world coordinates with center (wcx, wcy) and half-size ws and a square viewport in device coordinates with center (vcx, vcy) and half-size vs: (a) (b) (c) Show the basic transformations forming [T1] in symbolic functional form. Show [T1] as a single matrix. Find the expressions for the clipping coordinates of the world data point midway

between the window center and right boundary. (d) Show in symbolic functional form the basic transformations forming [T2] for

drawing on GRAFIC windows. 4. Window to viewport mapping transforms world data into device data based on a window (wcx,wcy,wsx,wsy) and a viewport (vcx,vcy,vsx,vsy). Derive this transformation as a single matrix with symbolic equations as its elements. Assume the X and Y data and device axes are parallel . 5. Draw a diagram illustrating endpoint codes for the line clipping algorithm. Explain how and why they are used. 6. Show the list of basic transformations in functional form that will make object B from object A in the following figure.
Y B A 1,1 (a,b) X 0,0

Given the rotated window and horizontal viewport, derive the window to viewport mapping equations, first in functional transformation form, then in matrix form.

Chapter 6. 2D Graphics Software

7.1 Window Graphics Systems

Chapter 7. Graphics Packages

During the late 1970s and early 1980s, raster graphics technology advanced rapidly and raster devices quickly replaced vector graphics devices. As a result of rapidly decreasing costs of computer memory and video technology, raster graphics systems became less expensive than vector systems. Most importantly, raster devices could economically provide color, which had become essential for many graphics applications. The software packages for vector systems were based on graphical segments. A segment was the organizational unit for graphics information and provided a logical structure for dynamic display programming. Using segments was like drawing on a stack of transparent pages held over the screen. The programmer could draw only on the top page but the stack could be shuffled to allow drawing on any page. Segments (pages) could be deleted, blinked, and even transformed individually. All displayable entities were placed in segments by the programmer. All segments appeared on the screen at once. It soon became clear that the segment-based graphics packages that had been used for vector display systems would not be appropriate for raster displays. Instead, the concept of a window was developed as the logical structure for organizing raster display systems.

7.1 Window Graphics Systems

The advent of workstations and personal computers greatly aided the development of window graphics systems. The basic window graphics systems in use today are described briefly in the next sections. The first two systems are kernel-based graphics systems, meaning the graphics functions and application must reside within the same computer. The last system described, the X Window System, is a network-based system, meaning the application and graphics functions can reside on different computers connected by a network.

Apple Macintosh Toolbox Apple Computer, Inc. developed a comprehensive suite of software for the Macintosh computer that first shipped in 1984. The software consists of a hundreds of functions that are placed

Chapter 7. Graphics Packages

7.1 Window Graphics Systems

in ROM (read-only memory) contained in each Macintosh [1]. The set of functions is called the Toolbox. A Macintosh program makes calls to Toolbox routines that are translated into system calls that enter the ROM code during execution.

Microsoft Windows Very similar to the Macintosh Toolbox, Microsoft Windows [2] was developed to support window graphics on IBM compatible personal computers. The first version of Windows appeared shortly after the Macintosh, and it has become the standard window environment for the IBM compatible world. Windows consists of a user environment that provides operating system functionality to the computer and contains graphics routines that are linked to programs executing Windows function calls. The routines support applications programming.

X Window System The X Window System [3] was developed jointly by Massachusetts Institute of Technologys Project Athena, Digitial Equipment Corporation and other companies starting in 1987. It was designed to support interactive raster graphics on networks of workstations. It has been adopted as a standard by nearly every workstation manufacturer. X Windows is network-based, meaning programs executing on one workstation calling X routines can display graphical data and interact with devices on another workstation or several other workstations. This makes it possible to run a program on a remote computer and interact with it on your local computer. This is termed a client-server model, where the client (application) program on a remote computer sends coded commands according to a protocol , or messages, over a network to a graphics server on the local computer. The commands are executed by the server to produce graphical output and other functions. The functionality of the X Window System is made available to programmers through Xlib, a library containing hundreds of C language routines for various functions. The Xlib routines are considered low-level functions, meaning they provide basic programming access to the protocol commands. Several software packages have been developed on top of Xlib routines to provide high-level functionality, for example, the OSF/Motif [4] user and programming environment. High-level refers to more sophisticated functionality intended for software developer support. There are some common elements in each of these window systems:

Chapter 7. Graphics Packages

7.2 Windows

windows

are the fundamental organizational structure for drawing on the screen and communicating with the user,

events

convey information about actions (user inputs, system happenings) to the program,

drawing primitives are lines, characters, fill regions, color and pixel operations. The following sections discuss these common elements.

7.2 Windows
Windows are in essence the raster counterpart to segments. They have a similar intent to segments, to allow a programmer to structure graphical display information in a manner appropriate for efficient and clear presentation and editing. Some of the first window systems were implemented as tile systems in which the tiles, usually rectangular areas on the screen, cannot overlap. In other words, the screen is divided into a number of non-overlapping regions, or tiled windows. Window systems relax this restriction and allow the regions to overlap.
Tiles Windows

Tile 1 Tile 3

Tile 2 Tile 4

Figure 7-1. Tiles and Windows. We can see that tile systems are considerably easier to implement. Tiles are a one-to-one mapping to screen pixels. Windows are a more difficult problem: 1. 2. Each pixel can belong to a number of windows (Figure 7-2). Only complex regions of windows may be visible. For example, the line a-b drawn in window A must be clipped against window B (Figure 7-3).

Chapter 7. Graphics Packages

7.2 Windows

A B

pixel is in all 3 windows C

Figure 7-2. Overlapping Windows and Pixels. A a B b C

Figure 7-3. Overlapping Windows and Clipping. 3. There must be means for ordering the windows. The stacking order of the previous figures, from the top , is B, C, A. Windows are not only an organizational convenience for implementation. They are useful for efficient graphical communication with a user. 1. A window is a metaphorical means for looking at the infinite world of data through a window. For example, scrolling, panning and zooming through a large picture or long document by moving the data through the window. 2. Windows are means for presenting many sets of data, or different views of one data set, to a user, like a stack of papers on a desk. The user can peruse the stack, move papers around as needed, pick one to edit, place them side-by-side for comparison, etc. 3. Dialog windows facilitate temporary communication with the user, such as warnings and data prompts, by re-using precious screen space.

Chapter 7. Graphics Packages

7.3 Events

7.3 Events
Graphics programs are notified of user actions, such as device inputs, and other noteworthy happenings through messages called events. Events control the execution of interactive programs and are central elements in all graphics systems. In general, all graphics programs are passive, waiting for events caused by actions to cause them to react. At the heart of each graphics system is the event processor that accumulates information and actions from devices connected to the computer and other sources, and places the data associated with the event (such as the key typed on the keyboard) at the end of the event queue. The typical graphics program contains at least one event loop that continually asks the question is there an event for me to process? In general, there is a function that removes an event from the head of the event queue and returns the data to the caller. For example, an event loop for a GRAFIC program is shown in Figure 7-4. The function GNextEvent returns four pieces of data: the for( ;; ) { /* forever */ eventId = GNextEvent( &wid, &a, &b, &c ); if( eventId == GKEYPRESS ) { if( a == a ) /* key a has been pressed on keyboard */ } if( eventId == GUPDATE ) { /* process window update event */ } } Figure 7-4. A GRAFIC Event Loop. window in which the event occurred (wid), and three data associated with the particular event (a,b,c). In the example, when a key is typed on the keyboard, an event of type GKEYPRESS is returned and the first argument (called a in the figure) will contain the ASCII code of the key. When running the X Windows version of GRAFIC, GNextEvent will call Xlibs XNextEvent; in Macintosh GRAFIC, the Toolbox function GetNextEvent is called; and in Microsoft Windows GRAFIC, PeekMessage is used in conjunction with a window callback procedure to accumulate events for GNextEvent. There are typically dozens of event types.

Chapter 7. Graphics Packages

7.4 Drawing Primitives

Finally, all graphics systems have a set of functions for drawing. As expected, at the heart of each package are basic functions for drawing lines and characters. Most packages have functions for other types of primitives, such as arcs, polygons (a closed loop of lines), rectangles, circles and ellipses, markers (a symbol at a point) and fill regions. Fill regions are closed areas bounded by lines (and possibly arcs) into which a color can be poured. Some systems provide sophisticated means for constructing fill regions, such as Boolean combinations (union, difference and intersection) of rectangles, circles, and other regions. Drawing primitives and windows are linked through coordinates. The coordinates for lines are given in window coordinates, Cartesian (x,y) values that are measured relative to the origin of a given window. Each window contains its own origin, and it is possible to draw in any window at any time. In GRAFIC (and all the window systems described previously), the window origin is its upper left corner and window coordinates extend right and down. Graphics systems also provide some representation for drawing attributes, such as color and drawing mode or function. Drawing mode refers to arithmetic and logical operations between the pixel value to be drawn and the corresponding pixel value in pixel memory. For example, think of drawing a line as an operation on the set of pixels in pixel memory where the line is to appear. For each of these pixels, the typical operation is to replace the pixel value in memory with the pixel value of the line. For example, drawing a red line means putting the pixel value of red into the pixels in pixel memory where the line is to appear. Now consider other operations than simply copying the pixel value into pixel memory. In GRAFIC, for example, a pen mode called GXOR means perform the Boolean exclusiveor function with the source pixel value, the pen color, and the destination pixel value, the corresponding pixel in pixel memory. This is illustrated in Figure 7-5. The first line drawn extends from (100,100) to (400,100) in black. The pen mode is changed to GXOR, and a second line is drawn from (200,100) to (300,100). For each pixel in the second line, the current color (black) is XORd with the corresponding pixel in pixel memory, in this case, also black. The result of black XOR black is white, as shown. Graphics packages typically provide a dozen or two such functions to control the drawing mode. Drawing attributes are usually maintained in a drawing state of some kind. When a drawing

Chapter 7. Graphics Packages

7.5 References

GPenColor( GBLACK GMoveTo( 100, 100 GLineTo( 400, 100 GPenMode( GXOR ); GMoveTo( 200, 100 GLineTo( 300, 100

); ); ); ); );

XOR Example

100

200

300

400

Figure 7-5. GRAFIC Calls Using XOR and Resulting Display. primitive is executed, its appearance is determined by the drawing attributes in the current drawing state. For example, in GRAFIC, each window has a separate drawing state that defines the color and drawing mode for the next line, character or fill region that is created. Functions are available for changing the drawing state, such as GPenColor to set the color, and GPenMode to set the mode. In the Macintosh Toolbox, the drawing state is called the GrafPort ; in Microsoft Windows, it is the device context; and in X Windows, it is the graphics context.

7.5 References
1. 2. 3. 4. Apple Computer, Inc., Inside Macintosh, Addison-Wesley Publishing Company, Inc., Reading, Massachusetts, 1985. Microsoft Inc., Microsoft Windows 95 Programmers Reference, Microsoft Press, Redmond, Washington, 1992. Nye, A., Xlib Programming Manual for Version 11, Volumes One and Two, OReilly &Associates, Inc., 1990. Open Sofware Foundation, OSF/Motif Users Guide, Prentice Hall, Englewood Cliffs, New Jersey, 1990.

Chapter 7. Graphics Packages

8.1 Pointing and Positioning

Chapter 8. Interactive Techniques

Recently, due largely to the rapid influx of personal computers in the home and workplace, considerable attention has been given to graphical user interfaces as an important aspect of programs and the key to ease of use. The design of a good graphical user interface, or GUI, is both an art and a science. The following sections describe some basic GUI techniques. The goals of a successful user interface should be to: 1. provide a natural person-machine interface, where natural means (a) appropriate and (b) efficient for the application, 2. 3. fully exploit the power of the machine to augment the users capabilities, let the machine do the work.

A good GUI should follow and exploit the principle of least astonishment: Make the expected happen; make what happens expected. Exploit consistency and user anticipation.

8.1 Pointing and Positioning

An important aspect of any GUI is the capability of pointing to parts of the visual presentation on the screen before the user, typically through the use of pointing and positioning devices coordinated with moving graphical entities on the screen. Cursors are one predominant form of pointing and positioning. A cursor is a graphical symbol, or icon, that moves on the screen according to user actions. The mouse and pointer symbol is a common example of this today. As the user moves the mouse with hand movements, a cursor moves on the screen. The user can move the cursor near other graphical entities on the screen and direct the program to perform some action. The form of the cursor can be an important source of information to the user. Special cursors can be used to convey problems or special input requirements as in Figure 8-1. As screen

Chapter 8. Interactive Techniques

8.1 Pointing and Positioning

position indicators, cursors can take different forms that are appropriate for certain tasks. For positioning busy error condition

text marker Figure 8-1. Examples of Special Cursors and Meanings.

example, cross-hair cursors are useful for aligning objects (Figure 8-2). cursor (x,y) screen or window horizontal line moves up and down with y cursor coordinate.

vertical line moves left and right with x cursor coordinate. Figure 8-2. Cross Hair Cursor. Positioning constraints provide more exact control of cursor motion according to userspecified options. Constraining the motion of objects connected to the cursor to either horizontal or vertical components of the cursor position facilitates accurate positioning operations (Figure 83). vertical constraint cursor unconstrained horizontal constraint

last point

Figure 8-3. Constrained Cursor Motion. Constraining the cursor location to fixed grid locations, sometimes called snap to grid, means the cursor is moved to the nearest imaginary (or visible) grid location that is computed for each cursor movement by rounding off the exact position to the nearest grid location (Figure 8-4).

Chapter 8. Interactive Techniques

8.2 Dragging

Figure 8-4.Cursor Positioning Constrained to a Grid.

8.2 Dragging
Feedback during positioning allows the user to preview the appearance of objects during re-sizing and moving operations. One of the most useful techniques for providing this feedback is to move objects with the cursor, called dragging. Dragging means drawing and erasing an object as its position is changed with the cursor position. With segment-based vector display systems, implementing dragging is relatively simple. The object to be dragged is placed in its own segment, distinct from segments displaying other data that will not change during dragging. As the cursor moves, the segment of the dragged object is continually erased and re-created, thereby creating the visual effect to the user of an object moving across the screen. In actuality, the object is not moving, it is an effect created by sequentially erasing the object drawn at the old position and drawing the object at a new position. Although the principle is the same, implementing dragging with raster graphics systems requires special operations because there is a problem with the erasing steps. At first thought, erasing can be implemented easily by drawing over the previously drawn object in the background color (typically white). Considering the object to be a straight line, the visual appearance of dragging is illustrated in Figure 8-5. This effect is called the rubber-band line. A first algorithm for the rubber-band line with a raster graphics package might have the following steps: 1. 2. 3. 4. 5. draw the line the from the start point to the old endpoint, get the new cursor location, called the new endpoint, redraw the old line, from start to old endpoint, in white to erase it, assign the new endpoint coordinates to the old endpoint, loop back to step 1.

The flaw in this approach is step 3, erasing the old line by drawing it in the background

Chapter 8. Interactive Techniques

8.2 Dragging

bounding rectangle

center (?) Figure 8-8. Characteristic Locations for Picking Text. value computations are computationally faster and create a rectangular picking region that is nearly indistinguishable to the user from the circular region. (Xc - Xi) 2+ ( Yc - Yi )2 < d 2

| Xc - Xi | < d AND | Yc - Yi | < d Figure 8-9. Picking Regions for Euclidean and Absolute Value Tests. Anywhere Along a Line: The problem is picking a bounded line (segment) with endpoints. We must combine the mathematical procedure for computing the distance to the unbounded line with the algorithmic process of limiting the picking region to the bounded portion of the line. The distance from a point to an unbounded line is classically computed as, Ax + By + C d = --------------------------------A2 + B2 where the line equation is Ax + By + C = 0, and A, B, and C are found from the line endpoints by (x,y) d (x2,y2)

(x1,y1) Figure 8-10. Distance from Point to a Line.

Chapter 8. Interactive Techniques

8.3 Picking

100

solving the simultaneous equations: Ax 1 + By 1 + C = 0 Ax 2 + By 2 + C = 0 Note that this involves dealing with a number of special conditions (A=0 or B=0, for example). Another method, based on a vector solution that will be discussed later, is: xl y yl x d = ------------------------------2 2 x + y where: x = x 2 x 1 y = y 2 y 1 l x = x x1 l y = y y1 Caution: the denominator above can be 0 when the line is actually a point. This test alone, however, has a picking region that extends infinitely along the line:

Figure 8-11. Picking Region for an Unbounded Line. To bound the picking region, perform tests according to the Boolean expression: picking region = inside the smallest rectangle surrounding the line and its endpoints AND within distance d of the unbounded line This is shown visually in Figure 8-12.

Chapter 8. Interactive Techniques

8.3 Picking

101

AND

Figure 8-12. Picking Regions for a Bounded Line.

8.3.3 Picking Circles

Picking a circle means that the cursor is within a given distance of either side of the circle, resulting in the annular picking region illustrated in Figure 8-13. One approach would be to YL (x,y) Yw R R+d XL (xc,yc) R-d Xw Figure 8-13. Picking Region for a Circle of Radius R and center (xc,yc). compute the distance from the cursor location (x,y) to the circle center (xc,yc) and compare this to the boundary radii of the two circles bounding the picking region, R+d and R-d. This approach is adequate for circles, but does not generalize for other conics (curves of degree two) such as ellipses. A more general approach is based on the properties of conic equations, i.e. in this case the equation of a circle: x2 + y2 - R2 = 0. Notice that if the point (x,y) is inside the circle then x2 + y2 R2 < 0 and if the point (x,y) is outside the circle then x2 + y2 - R2 > 0. To simplify and speed the computations, first transform the cursor location into the local coordinate system with its origin at the center of the circle. This transformation is simply (x,y) - (xc,yc). We now test the location of

Chapter 8. Interactive Techniques

8.3 Picking

102

this point with respect to the picking region, which can be expressed logically as: inside the outer circle (radius R+d) AND outside the inner circle (radius R-d), or more conveniently, inside the outer circle (radius R+d) AND NOT inside the inner circle (radius R-d). The second form simplifies the logic to one routine Boolean InsideCircle( x, y, radius ).

8.3.4 Picking Composite Objects

A composite object is an object that consists of a collection of simpler objects, such as lines, circles, or other entities. Picking composite objects is application dependent, meaning the user may or may not have characteristic locations for objects that are natural picking points or regions. Vertices are often convenient, except for curved objects.

Curves have no vertices!

Figure 8-14. Vertex Picking Examples. Picking an object that is composed of lines means picking any of its lines.

Figure 8-15. Picking an Object Composed of Lines. Picking an object by picking near its centroid, i.e., the geometric center of an object, can sometimes be useful (but not often). A bounding rectangle is the most often used picking region. This is visually understandable but also can be a gross over-approximation to the inside of an object. Picking by number or name should be used only when numbers or names are inherent to

Chapter 8. Interactive Techniques

8.3 Picking

103

Figure 8-16. Centroids of Objects.

picking region much larger than object Figure 8-17. Bounding Rectangles of Objects. the application, i.e. objects have names known by the user anyway. Even in this case, graphical picking should be permitted too.
Fred 12

Figure 8-18. Showing Object Names. Picking inside objects is a sophisticated approach to picking that is useful for simple objects, such as rectangles, but is very complicated for other objects, especially those with curves.

Figure 8-19. Interiors of Simple and Complex Objects. Inside/outside tests, or point containment tests, will be discussed in a later chapter. Picking inside a selection region is useful for selecting all objects in an area, usually a rectangle, on the screen. This requires special tests for the intersection of the selection region with the object

Chapter 8. Interactive Techniques

8.3 Picking

104

start with button down here

TEXT

drag mouse with button down

release button here Figure 8-20. Picking Objects Inside a Selection Region. region, either the bounding box, individual lines, or vertices.

8.3.5 Interaction Paradigms

Verb - Noun Paradigm The interaction model in most early interactive graphics systems can be characterized as verb-noun paradigm because the user first selected the action (verb) and then was prompted for the entities on which to act (nouns). Nouns are selected objects, such as text characters, graphical objects, markers, etc., whatever objects are pertinent to the application. Verbs are commands that operate on objects. This command structure approach usually required hierarchical menus and numerous prompts to request picks from the user. Research in user interfaces showed that such systems are very difficult to learn, difficult to remember and require an excessive number of user inputs. One of the main drawbacks of the verb-noun paradigm is that the user (and program) is continually entering a mode, meaning a program state requiring special user inputs to complete. Modes usually require additional user inputs (e.g. enter mode, exit mode) and tend to challenge users to recall where they are and what is required to exit and complete the mode. In addition, verbnoun systems must handle many error conditions when the user picks invalid data items.

Noun-Verb Paradigm Recent systems, first popularized by the Apple Macintosh graphical user interface, are based on a noun-verb paradigm. At first glance, the noun-verb paradigm appears to be a simple

Chapter 8. Interactive Techniques

8.4 Review Questions

105

reversal of picks and commands, but in practice it greatly simplifies user interfaces. The user selects (and un-selects) objects (the nouns), and then executes the desired command (verb). Noun-verb systems require that the program and user maintain a selection, meaning a set of selected objects. The selection is highlighted to visually distinguish selected objects from unselected objects. For example, selected text is drawn inverted (or outlined by a darkened box, etc.), or a rectangle has tabs called handles drawn at its corners. Knowing the selection before executing the command allows the program to: 1. enable and disable menu items and change their text to be appropriate for the selected objects and more informative to the user, 2. change the action performed to depend upon the selected objects.

For example, a delete command can be disabled (grayed and not allowed to be executed by the user) when no object is selected, different commands can be enabled when different numbers or types of objects are selected, and so on.

8.4 Review Questions

1. Explain an algorithm for picking a transformed circle. Assume the circle is transformed by a matrix M as it is drawn, and that the user picks a point on the display. 2. Summarize the verb-noun and noun-verb paradigms for graphical user interfaces. Give an example of how a delete object command could be changed to follow the verb-noun paradigm.

Chapter 8. Interactive Techniques

106

Chapter 9. Data Structures for Interactive Graphics

As you have probably seen to this point, graphics programs are always operating on data. In a non-graphical program, there is only one use for the data: analysis. By incorporating interactive graphics, data is now also used for interaction and display. Accordingly, we have two types or categories of data: 1. Problem data: for analysis (e.g., motion analysis of a mechanical device requires the locations of joints and dimensions of links) 2. Graphical data: for display (e.g., creating and displaying the mechanical device on the screen) In general, these types of data are not always distinct, i.e. there is overlap where data is used for both (Figure 9-1): Data Geometry

problem

graphic

Display

Analysis Figure 9-1. Problem and Graphical Data. Examples: Graphical Data Geometry Visual Properties (color) Graphics Package Information (windows, etc.) The following development of data structures combines methods for representation of objects with the needs for graphical interaction. We will use the application idea of sketching Problem Data Geometry Physical Properties (density)

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

107

objects consisting of lines as a basis for discussion of both the data structures and interaction.

9.1 Basic Data Storage Methods

First examine the data needed to create one object, one entity at a time. In our sketching application, points and straight lines are the entities, but the ideas can be generalized to other geometric and non-geometric forms. Keep in mind that the controlling program is interacting with a user whose commands activate and direct the actual storage and manipulation of the data, in this case lines. As the object is drawn on the screen, which is purely a graphical operation of the program, it is also necessary to store the lines as data for later reference, either for analysis ( What is the length of the line? ), or for editing operations ( Delete this line ).

9.1.1 Compacted Sequential Lists

The most direct method of storing data is a compacted sequential list, or CSL, as illustrated in Figure 9-2. data[] start free 0 1 2 1st entity 2nd entity unused ( free )

MAX-1

unused ( free )

Figure 9-2. Compacted Sequential List Storage. The variables start and free are pointers that indicate locations in the data that hold the first used entity and first free cell, respectively. In a CSL, start always points to the first cell, index 0, and is actually unneeded. In a CSL, entities are first, followed by free cells, in two contiguous blocks of cells.

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

108

The entities can be stored in several array(s) indexed by the same variable. For example, if one were storing a line drawing as a sequence of endpoints marked by pen codes, there would be three arrays (see Figure 9-3).

x[] x1 x2 free x3 y1 y2 y3

y[]

pen[] MOVE DRAW DRAW x1,y1 x2,y2 x3,y3

In the CSL storage method, adjacency between array positions was used to determine which points were connected to form lines, and in general how to find the previous and next entities in the data. Thus, connectivity is implicit in a CSL. Another method for storing data is a linked list. It appears similar to sequential storage, except the next entity need not follow the current one in storage. In other words, the data is linked through explicit pointers stored with each entity, as shown in Figure 9-9. Start points to the first entity (not necessarily the first cell in the data) of the entity list that defines the object. Free points to the first cell of the free cell list. Lists are terminated by a NULL pointer, a unique pointer value (often, but not always, 0). The display on the screen is the same, only the internal storage method has changed. An array of integers can be added to hold the next pointers for the multiple array method. With structures, an int next; field is added to the definition of an entity. In the next section, we will use a C pointer variable for next. It usually is easier to think of, and draw, a linked list as a graph of boxes connected by

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

112

Entity data[] start free 0 1 2 unused ( free ) 1st entity x1,y1,pen1 unused ( free ) 1st entitys next pointer 2nd entitys next pointer is NULL, for end of list.

MAX

2nd entity x2,y2,pen2

Figure 9-9. Linked List Storage. arrows as shown in Figure 9-10.

start

x1,y1,pen1

x2,y2,pen2

free Figure 9-10. Diagram Form of Linked List. The boxes are entities or cells, and the arrows are the links. The arrow with a line is a NULL pointer. Now consider the basic operations as before, using the same example as before. 1. Initialization: Initialization now means: set start to NULL and link all cells into the free list.

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

113

data[] start free 0 1 2 unused ( free ) unused ( free ) unused ( free )

MAX Figure 9-11. Initialization of a Linked List Structure. One possible initialization loop is illustrated in Figure 9-12.
start = NULL; for( i = 0; i < MAX-1; i++ ) data[i].next = i + 1; data[MAX-1].next = NULL; free = 0;

Figure 9-12. Linked List Initialization. 2. Appending an entity: The three steps are the same: (1) get a free cell, (2) append the cell to the object, (3) fill the cell with the entity data. We now have an end condition to consider: when start is NULL, there is no last entity on which to append. Figure 9-13 illustrates one approach to appending to a linkedlist. It may be more convenient and efficient to maintain a variable last that always points to the end of the list. 3. Deleting an entity: Deletion shows the real advantage of linked lists for applications requiring dynamic data structures, i.e. data structures that are created, deleted, and edited interactively. Assume we have picked the entity to be deleted from the CSL and recorded a pointer to it in the variable here. The pointer prev indicates the previous entity, i.e., the entity that points to here. Note that the first entity will not have a prev entity, so we handle this specially as another end condition. Given start, we can find prev as shown in Figure 9-15. Now we are set to unlink the entity and return it to the free list as shown in Figure 9-16. The free list is generally unordered, so returning the cell to the front is the easiest method.

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

114

/* step 1: get a free cell */ new = free; if( new == NULL ) Error(OUT OF SPACE ); free = data[free].next; /* step 2: find the last entity in the entity list */ if( start == NULL ) { start = new; last = NULL; } else { last = start; while( data[last].next != NULL ) last = data[last].next; } if( last != NULL )

data[last].next = new;
/* step 3: */ data[new].x = newx; data[new].y = newy; data[new].pen = newpen; data[new].next = NULL; /* dont forget to init. next! */

Figure 9-13. Appending An Entity to a Linked List. prev here

ENTITY A

ENTITY B

ENTITY C

Figure 9-14. Linked List Deletion.

if( here == START ) prev = NULL; else { prev = start; while( data[prev].next != here ) prev = data[prev].next; }

Figure 9-15. Finding the prev pointer.

/* first unlink the entity: */ if( prev == NULL ) start = data[here].next; else data[prev].next = data[here].next; /* return the freed cell to the front of the free list */ data[here].next = free; free = here;

Figure 9-16. Unlinking the entity referenced by here. 4. Inserting an entity:

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

115

Inserting an entity is similar to appending one, so much so in fact that they can be combined without too much difficulty. Assume that we have already obtained a free cell and stored the data in the entity pointed to by new as illustrated in Figure 9-17. new

entity new entity B

entity A

Figure 9-17. Inserting a new entity. We use prev==NULL to mean at the front. The procedure can be coded as shown in Figure 9-18:.
/* link the entity into the object after the entity prev */ if( prev == NULL ) { data[new].next = start; start = new; } else { data[new].next = data[prev].next; data[prev].next = new; }

Figure 9-18. Code to insert a new entity.

9.1.3 Linked Lists with C/C++ Pointers

Instead of integer pointers, it is more efficient and helps modularity to use C/C++ pointer variables (or Java class references) to represent the start and next pointers in the Entity data. In the C/C++ case, the Entity declaration could be: typedef struct int int Entity* } Entity; Entity* start; Entity* free; entity {/* or class Entity { in C++ */ x, y; pen; next; /* pointer to next Entity */ /* start pointer */ /* free pointer */

Chapter 9. Data Structures for Interactive Graphics

9.1 Basic Data Storage Methods

116

(The line Entity* next will not compile as shown, but is written this way for easier reading.) In the next figures, the previous code fragments are re-written to use the next pointer. Figure 9-19 shows initialization using pointers, Figure 9-20 shows appending an entity, deleting
start = NULL; /* NULL is usually 0 for C/C++ pointers */ for( i = 0; i < MAX-1; i++ ) data[i].next = &data[i + 1]; data[MAX-1].next = NULL; free = &data[0];

Figure 9-19. Linked List Initialization Using Pointers.

/* step 1: get a free cell */ Entity *last, *new; new = free; if( new == NULL ) Error(OUT OF SPACE ); free = free->next; /* step 2: find the last entity in the entity list */ if( start == NULL ) { start = new; last = NULL; } else { last = start; while( last->next != NULL ) last = last->next; } if( last .NE. NULL ) last->next = new /* step 3: */ new->x = newx; new->y = newy; new->pen = newpen; new->next = NULL; /* dont forget to init. next! */

Figure 9-20. Appending An Entity to a Linked List Using Pointers. an entity using pointers is shown in Figure 9-21 and inserting an entity is shown in Figure 9-22.
if( here == START ) prev = NULL; else { prev = start; while( prev->next != here ) prev = prev->next; }

Figure 9-21. Finding prev And Unlinking An Entity Using Pointers.

Chapter 9. Data Structures for Interactive Graphics

9.2 Storing Multiple Objects

117

/* first unlink the entity: */ if( prev == NULL ) start = here->next; else prev->next = here->next; /* return the freed cell to the front of the free list */ here->next = free; free = here;

Figure 9-21. Finding prev And Unlinking An Entity Using Pointers.

/* link the entity into the object after the entity prev */ if( prev == NULL ) { new->next = start; start = new; } else { new->next = prev->next; prev->next = new; }

Figure 9-22. Inserting a New Entity Using Pointers.

9.1.4 Using Compacted Sequential Lists and Linked Lists

Linked lists require more storage than CSLs. CSLs, however, require the movement of entities for insertion and deletion. If the entities consist of much data, and/or if there are large numbers of entities, CSLs can be slow. Also, in more complex data structures (as will be discussed next) where entities may be referenced by other entities, moving an entity requires that its reference(s) be updated accordingly. This can be clumsy, so linked lists would be preferred.

9.2 Storing Multiple Objects

Many applications operate on more than one object. This requires that a number of objects be available to construct a scene. Objects are groups of entities created by the user. Now we must maintain groups of objects, or groups of groups of entities. Consider our basic storage techniques for representing groups of entities. We define the structure Object to contain a start and end pointer: typedef struct { int start; int end; } Object; /* Entity* start for C/C++ pointers */ /* Entity* end for C/C++ pointers */

Chapter 9. Data Structures for Interactive Graphics

9.2 Storing Multiple Objects

118

Object objects[MAXOBJ]; A second data structure, the object table, in this case a CSL, contains pointers to the start and end of each object in the Entity data. For an Entity data structure that is a linked list, the object table needs only the start pointer and can appear as shown in Figure 9-23. All that is needed is object table: entity storage:
entity[]

object 1 object 2 freeobj MAXOBJ

start

free

Figure 9-23. Object Table Using Linked Lists for Entity Data. the pointer to the first entity, called the head, of a list. The object table is simply a CSL of list head pointers.

Chapter 9. Data Structures for Interactive Graphics

9.3 Instancing with Multiple Objects

119

9.3 Instancing with Multiple Objects

It is likely that what have been called objects would be used as basic definitions (or building blocks) to create a scene. There may be two separate programs or two modes in one program, one to create the objects and another to create the scene given the object definitions. It is wasteful and usually undesirable to duplicate the list of entities (no matter how they are stored) defining an object every time it is added to the scene. Instead, we add another level of data to define the scene in terms of instances of the previously defined basic definitions. An instance is a reference to a pre-defined object varying in position, size, orientation and attributes. Note that each instance of a given object references the same entity data. Consider the objects and scene shown in Figure 9-24. In addition to the object table and

The Scene: Objects: 2 box triangle

3 1

Figure 9-24. Instances of Basic Objects in a Scene. entity data, another data structure is needed to represent the instances of objects in the scene, an instance table, which stores data for each instance. For this, we could define a structure and type called Instance: typedef struct { Object* object; /* pointer to basic object */ float tx, ty; /* the translation */ float sx, sy; /* the scale factors */ TPixel color; /* Grafics color of the instance */ /* etc. */ } Instance; Instance instances[MAXINSTANCE]; The instance table for the scene in Figure 9-24 is illustrated in Figure 9-25. The instance table contains a reference to the object, i.e., a pointer to the object table or a string that is the name of the object that can be matched with the name stored in the object table. An integer array index

Chapter 9. Data Structures for Interactive Graphics

9.3 Instancing with Multiple Objects

120

for object is shown above, but a C/C++ pointer could be used as well as illustrated in previous sections. The instance table also contains attributes for each instance, which can be any geometric, visual or other properties that modify the appearance, location, scale, rotation, etc., of one instance of an object. Each instance is a transformed copy of a basic object. One instance of a basic object has the same basic definition as another, i.e. each instance of a box contains the same entities (lines). Each instance, however, has a different appearance on the screen due to its different location, size, color, etc. Often, the location, size and orientation of an instance are represented in a transformation matrix using the basic transformations of translation, scaling and rotation, respectively. instance table: Instance instances[MAXINSTANCE] 0 1 2 box ,tx1,ty1,sx1,sy1,color1 box ,tx2,ty2,sx2,sy2,color2 box ,tx3,ty3,sx3,sy3,color3

3 triangle ,tx4,ty4,sx4,sy4,color4 object table: Object objects[MAXOBJ] Entity data[MAX]: x,y,pen box , start triangle , start

Figure 9-25. Instance Data Structures. The parameters in the instance table modify the appearance of the basic object when this instance is drawn. The object is transformed geometrically and visually as it is drawn. For this example, the instance drawing loop is outlined in Figure 9-26. The coordinates are transformed by the instance transformation representing the location, size and orientation, and other visual properties can sometimes be implemented through simple graphics package calls. The instance table can be represented using a CSL or linked list structure,

Chapter 9. Data Structures for Interactive Graphics

9.4 Hierarchical Modeling with Instances

121

for( i = 0; i < #instances; i = i + 1 ) { GPenColor( instances[i].color ); BuildTransform([M],instances[i].tx,instances[i].ty, instances[i].sx[i], instances[i].sy ); for( each point j of object[i] ) { Transform (data[j].x,data[j].y) by [M] into (x,y); if( data[j].pen[j] == MOVE ) GMoveTo( x, y ); else GLineTo( x, y ); } }

Figure 9-26. An Example Loop to Draw Instances Using Grafic. although the CSL is adequate for this example. Picking objects on the screen means picking the instance of an object, which in turn means picking a transformed copy of an object. Thus, the cursor coordinates must be compared to the transformed coordinates of each particular instance of an object. Another approach that is more computationally efficient is to transform the screen coordinates of the cursor into basic definition coordinates of each instance and then to compare these coordinates to the basic object coordinates. This involves multiplying the cursor coordinates by the inverse of the instance transformation of the object in question.

9.4 Hierarchical Modeling with Instances

To this point in the discussion there has been only one level of data. No instance is related in any way to another; each is simply in the scene. This may not be sufficient. For example, to reposition the house in the previous scene each entity must be moved individually. The data structure has no concept of house. Another example is an airplane, with engines connected to wings connected to a fuselage. To properly represent and manipulate this data requires a hierarchical scene definition. Extend the previous instance data structure with relational information, four pointers called prev (for previous), next, parent and child, that point to other instances in the instance table, that in turn point to four other instances. A graphical depiction of this structure is shown in Figure 9-27 where pointers are shown as horizontal and vertical arrows. Each box is an instance data node that contains the basic object reference and the geometric attributes, often as a

Chapter 9. Data Structures for Interactive Graphics

9.4 Hierarchical Modeling with Instances

122

Data Node parent prev Ptr to Data & Transform child Figure 9-27. Graphical Depiction of Hierarchical Relationships between Entities. transformation matrix. In this representation, special significance is given to the parent-child relations. They represent a hierarchical relationship between instances that relates the position, size and orientation of each child relative to its parent. The prev and next pointers are a doubly-linked list of children of a parent that has no such geometric significance. Figure 9-28 is an example of how an airplane could be modeled with such a structure. next

Airplane

Fuselage

Left Wing

Right Wing

Outboard Engine Left

Inboard Engine Left

Outboard Engine Right

Inboard Engine Right

Figure 9-28. Airplane Example of Hierarchical Data Structure.

Chapter 9. Data Structures for Interactive Graphics

9.4 Hierarchical Modeling with Instances

123

What is the transformation in the data node? There are two possibilities: absolute (global) or relative (local). An absolute transformation maps the data from local coordinates to world coordinates. The airplane flies in world coordinates. Using absolute transformations, when the transformation of any node is changed, all children must be changed appropriately. This is awkward and somewhat defeats the hierarchical nature of the representation because there is really no significance to the parent-child relation. In essence, with absolute transformations each node is independent of the others. A relative (local) transformation maps the childs coordinates to its parents coordinates. That is, each child transformation is relative to its untransformed parent. This localizes the transforms and makes the representation significantly more flexible. Transforming the airplane now requires changing only one matrix (and re-drawing the airplane, of course). The computation of the absolute transformation for each node is done now as part of the drawing process. Using relative transformations, the location, size and orientation of a wing that is defined in its own convenient coordinate system is defined by the location, size and orientation of the fuselage to which it is attached (i.e., its parent). Therefore, when traversing this hierarchical data structure, for example for drawing, it must be understood that transforming a node (data and associated transformation) in the tree also transforms all of its children. In other words, move the plane, and everything connected to the plane moves too. The parent-child relation is now a geometric association. This hierarchical data structure is greatly aided by transformation matrices. For example, a division of a company has many designers working long hours developing the data for a wing. The designers naturally adopt a convenient coordinate reference frame in which to express the wing data, which we can consider for now as a list of points and pen codes. Similarly, another division of the company has developed the fuselage data and another the engine data, likewise selecting convenient reference frames (different than the others). Now, it is time to develop a model of the airplane with two identical wings (well use mirror images for our 2D graphical example, but this would not be done with 3D data) connected to the fuselage, each wing with two identical engines. It would be unreasonable (and unnecessary) to ask one division to recompute their data in the reference frame of the other. Instead, we need to relate the reference frames, and use instance transformations to model the airplane structure as a hierarchy. The wing data defined in a wing reference frame (coordinate system) , WW , is first

Chapter 9. Data Structures for Interactive Graphics

9.5 The Current Transformation Matrix

124

positioned relative to a fuselage defined in its reference frame, FF, by a transformation MF/W (read as transformation from wing to fuselage ). Similarly, an engine defined in an engine coordinate frame, EE, is positioned relative to the wing by MW/E. The coordinates of the wing expressed in the fuselage coordinate system WF are: WF = WW * MF/W The coordinates of the engine expressed in the wing coordinate system EW are: EW = EE * MW/E And now the engine can be expressed in the fuselage system as EF: EF = EW * MF/W = ( EE * MW/E ) * MF/W Notice that EF = EE * ( MW/E * MF/W ) or EE * MF/E where MF/E = MW/E * MF/W . This shows that as we descend from parent to child in the hierarchy, we can concatenate matrices to accumulate the global transformation. This concatenation process continues as we further descend the structure. Consider, however, the transformations of the children of one parent, such as the fuselage. Transformations of one child do not affect transformations of other children with the same parent. Consider the example illustrated in Figure 9-29. To compute the transformation relating the outboard engine left to the airplane, MOELA, we must pre-concatenate the transformations of each parent, from airplane to outboard engine left: MOELA = MOEL * MLW * MF * MA

9.5 The Current Transformation Matrix

An efficient method for maintaining the transformations while traversing a hierarchical data structure is to use a current transformation matrix, called CTM. The CTM holds the accumulated global transformations for the current node and all its parents. When descending the data structure, parent to child, transformations are PRE-concatenated with the CTM, i.e. CTM = Mchild * CTM (for row vectors). The child node is then drawn with

Chapter 9. Data Structures for Interactive Graphics

9.5 The Current Transformation Matrix

125

Yf Xf Yw MA = anything Xw Ye Xe MF = R(90) Airplane

Ya Xa

C AB

Fuselage

Left Wing MLW=S(-1,1)R(-90)T(0,A) MRW=R(-90)T(0,-A)

Right Wing

Outboard Engine Left MOEL=T(C,0)

Inboard Engine Left MIEL=T(B,0)

Outboard Engine Right MOER=T(C,0)

Inboard Engine Right MIER=T(B,0)

Figure 9-29. Numerical Airplane Hierarchy Example. the current CTM: P' = P * CTM. (NOTE: this is for row vectors. For column vectors, the transformations and the order must be transposed.) However, examine the CTM for the Left Wing versus the Right Wing of the Fuselage. After drawing one child of a parent, the starting CTM for the next child is the parents CTM. This process repeats as the child becomes the parent, and so on. If we wish to maintain the prior parents CTM as we display each parent-child-child in a depth-first manner, we must store all the prior CTMs in a LIFO (last-in first-out) CTM stack. A stack is simply an array structure that has a top element index, or stack pointer. The stack entries, in this case, are transformation matrices. This can be implemented using a three dimensional array CTMSTACK[i,j,k]. Most systems maintain a stack of transformations and use the top transformation as the CTM. Prior to concatenating a childs transformation with the CTM, we push the parents CTM onto the stack. This is done by incrementing the stack pointer and copying the CTM into the

Chapter 9. Data Structures for Interactive Graphics

9.5 The Current Transformation Matrix

126

stack entity at this pointer. The CTM is now a copy of the previous CTM, which is the next transformation on the stack. Subsequent transformations will concatenate to the new CTM. After the child has been drawn with its CTM, we must restore the prior CTM before drawing the next child by popping the top transformation off the stack and restoring the previous CTM as the top of the stack. This is done simply by decrementing the stack pointer. Figure 9-30 illustrates CTM tracing code for this hierarchical data structure.
void DrawObject ( ObjectPointer object, TransformMatrix ctm ) { TransformMatrix newctm; /* new matrix allocated each call */ ObjectPointer child; MatrixMultiply: newctm object->xform ctm DrawData( object->data, newctm ); for(child=object->child; child != NULL; child=child->next ) DrawObject( child, newctm ); }

Figure 9-30. Hierarchical Drawing Routine. The execution of the routine in Figure 9-30 using the data in Figure 9-31 is traced in Figure

C
Figure 9-31. Example Hierarchy. 9-32.

Chapter 9. Data Structures for Interactive Graphics

9.6 Modeling Languages and Scene Graphs

127

DrawObject( A, I ) newctm1 = MA I DrawData( A, newctm1 ) for( child = B ) DrawObject( B, newctm1 ) newctm2 = MB newctm1 DrawData( B, newctm2 ) for( child = C ) DrawObject( C, newctm2 ) newctm3 = MC newctm2 DrawData( C, newctm3 ) ... for( child = D ) DrawObject( D, newctm1 ) newctm4 = MD newctm1 DrawData( D, newctm4 )

Figure 9-32. Trace of Execution DrawObject Routine.

9.6 Modeling Languages and Scene Graphs

Hierachical modeling with transformations is a fundamental aspect of most graphics systems. This will be especially important in three dimensional modeling. One example of the use of instances and transformation is the graphics modeling language introduced in the 1980s for the Evans & Sutherland Corporations PS300 3D graphics system. Figure 9-33 is an example of a PS300 program. . The program statements describe lines, transformations and instances of graphical objects that are created using named structures. The diagram in the lower right corner of Figure 9-33 is a graphical depiction of the program in which each graphical object is a node drawn as a circle and the arrows represent the instance statements calling other objects. The caller is the parent and the called node is the child. The system automatically pushes the CTM before an instance call and pops the CTM afterwards. The PS300 modeling language is a pre-cursor in some ways to the VRML (Virtual Reality Modeling Language) statements and structure. Both languages use a hierarchical (parent-child) structure with transformations to model objects and geometric relations. In modeling languages, the statements represent a scene as a graph of nodes and relations, similar to, but more general than, the previous hierarchical data structure discussion.

Chapter 9. Data Structures for Interactive Graphics

9.6 Modeling Languages and Scene Graphs

128

Yb "box" box := 0, 0.25, 0.25, 0, 0, VECTOR_LIST 0 0 0.25 0.25 0; N=5 Xb Yh "house"

roof := BEGIN_STRUCTURE TRANSLATE BY -.05,.25; SCALE BY 1.4,.2; INSTANCE OF box; END_STRUCTURE; door := BEGIN_STRUCTURE TRANSLATE BY .12,0; SCALE BY .2,.8; INSTANCE OF box; END_STRUCTURE; house := BEGIN_STRUCTURE INSTANCE OF box; INSTANCE OF roof; INSTANCE OF door; END_STRUCTURE; houses := BEGIN_STRUCTURE INSTANCE OF house; TRANSLATE BY .4,0; INSTANCE OF house; END_STRUCTURE;

Xh Yhs "houses"

Xhs

houses

house

box

roof

door

etc.

box

Figure 9-33. The Evans & Sutherland PS300 Hierarchical Modeling Language. Such representations have been termed scene graphs. A recent graphics system based on scene graphs is Java3D by Sun Microsystems. Java3D is a set of Java classes that provides sophisticated and efficient 3D graphics support for constructing a scene graph dynamically. The scene graph is a tree of objects descending from the root node called the virtual universe. The system does all processing, and drawing, from the scene graph data.

Chapter 9. Data Structures for Interactive Graphics

9.7 Review Questions

129

9.7 Review Questions

1. Assume a program has displayed the hierarchical object above in a GRAFIC window using window to viewport mapping. The window surrounds the data and the viewport is the entire GRAFIC window. Describe, in step-by-step form, a picking algorithm for selecting an instance on the screen, given the cursor's viewport coordinates. Use the "first within tolerance" method. Show and explain all necessary equations and computations. 2. A hierarchical 2D data structure consists of a parent and two child instances of the same basic object. Show the relative transformations Tp, T1 and T2,in functional form. There are no scale transformations. basic object world coordinates YW YL 1 G 1 XL child2 G G XW G child1 3. Give the matrix equation of basic transformations that transforms a unit circle at the origin into an ellipse with major axis a, minor axis b, center (cx,cy) and orientation theta from the +x direction.
Chapter 9. Data Structures for Interactive Graphics

hierarchical data structure parent Tp

parent child1 T1 child2 T2

130

Chapter 10. 3D Rendering

Three-dimensional (3D) coordinates can be stored easily as data simply by adding the z coordinate. The problem is creating a two-dimensional (2D) image of a 3D scene. There are three types of such pictorial renderings: 1. Wire Frame. The data is drawn by connecting points with lines or curves. This is the

simplest computationally, but is spatially ambiguous. Can you see two different views in Figure 10-1? A Is line C-D closer or farther away than line A-B? C B Figure 10-1. Ambiguity in Wire Frame Images. 2. Hidden Lines Removed (HLR). Only parts of lines or curves that are not covered by other D

faces (surfaces) are drawn. HLR involves more complicated and time-consuming computations, but produces an image with less ambiguity. A Is point A closer or farther away than B? B Figure 10-2. Figure 10-1 with Hidden Lines Removed. 3. Hidden Surfaces Removed (HSR). HSR output requires raster displays. Those parts of

faces not hidden by other faces are shaded (filled with pixels) with colors that indicate the color and intensity of light reflecting from the face. This involves the most complex and time-consuming computations, and produces the least ambiguous images. Special lighting effects include reflection, refraction and shadows to make the image more realistic.

Chapter 10. 3D Rendering

10.1 Depth Cueing

131

Shading is based on the intensity of light reflecting from the faces and indicates their relative angle to the eye.

Figure 10-3. Hidden Surface Image of Figure 10-1.

10.1 Depth Cueing

The basic problem in 3D rendering is to create the illusion of depth in the 2D image. Several techniques can be used singly or in combination to help remove ambiguity from the 2D image, especially for wire frame images. 1. Perspective projection causes 3D parallel lines to converge in 2D to a vanishing point, like

the classic railroad tracks pictures. vanishing points

Figure 10-4. Perspective Image Showing Converging Lines. Images that do not use perspective projection, that is, those in which all parallel lines in 3D remain parallel in 2D, are termed orthographic projection. The most common use of orthographic projection, the three-view drawing, in essence ignores one coordinate when drawing each 2D orthographic view (Figure 10-5).

Chapter 10. 3D Rendering

10.2 3D Data Structures

132

Top Top Side

Front X Front Z
X Z Y Y

Side

Figure 10-5. Orthographic Images in Engineering Drawing Arrangement. 2. Intensity Depth Cueing displays lines farther away from the eye at lower intensities. This

can be done relatively easily in software for constant intensity lines on raster displays. Some high performance display systems offer this as an option. 3. Stereoscopic Viewing requires two projected views, one computed for the left eye and one

for the right. Special viewing apparatus is often required. Color can also be used to cue the two images, such as the red and green lens stereo glasses worn in movie theaters. 4. Kinetic Depth revolves the object dynamically, causing lines farther away to move more

than lines nearer to the eye. This requires dynamic 3D rotation, which is available on many display devices today (for additional cost, of course).

10.2 3D Data Structures

Points, curves and surfaces are the basic elements of 3D geometry. In general, considering only the mathematical forms, curves and surfaces are unbounded, meaning they have no start or end. Lines and planes extend infinitely. Practically, 3D geometry is bounded. A line has endpoints and a plane is bounded, i.e. has sides or boundaries. The terms vertex, edge and face are used to describe the topological elements of 3D objects related to points, curves and surfaces of geometry, respectively. A face is a surface bounded by edges, an edge is a curve bounded by vertices, and a vertex is a point on an object. For example, the cube in Figure 10-6 consists of topological elements of eight vertices, 12 edges bounded by these vertices, and 6 faces bounded by these edges.

Chapter 10. 3D Rendering

10.2 3D Data Structures

133

Geometrically, there are 12 points, each edge is a curve that is a straight line, each face is a planar surface. As another example, a cylinder has no vertices, two circular edges, and three faces, two planar (bounded by the edges) and one cylindrical (bounded by the edges). E B F A

H C

G D

Figure 10-6. A Cube and its Eight Vertices. Pictorial representations require different data storage methods. A box, for example, with vertices named A-G, as shown in Figure 10-6, can be stored several ways. The type of storage method can be dictated by the type of rendering(s) desired.

Wire Frame Representations This is the lowest level of the data storage methods. Only curves (lines) are needed, so any of the storage methods described in the previous chapter can be used. For the cube in Figure 10-6, the data structure would store the 12 lines: AB, BC, CD, DA, EF, FG, GH, HE, BE, CH, FA, GD. One approach to storing lines would be to store the two endpoints of each line, i.e., (x1,y1,z1) and (x2,y2,z2). Notice that this duplicates each point three times (three lines meet at a point in a cube) which can cause redundant calculations when transforming the object. Another storage method, called points and lines, stores points and lines separately. A line consists of references to its two endpoints, analogous to saying the line starts at vertex i and ends at vertex j. The points and lines representation of the cube is shown in Figure 10-7. Notice that each vertex is stored once, affording efficiency in space and computation time, because transforming the object involves transforming each vertex only once. In general, edges can be other than straight lines.This would make it necessary to store the type of curve or its equation with each edge. The edge is still a path between vertices, but now the description of this path, the curve, would be explicitly given.

Chapter 10. 3D Rendering

10.2 3D Data Structures

134

index 1 2 3 4 5 6 7 8

points x[] y[] xA yA xB yB xC yC xD yD xE yE xF yF xG yG xH yH

z[] zA zB zC zD zE zF zG zH

lines index Start[] End[] 1 1 2 2 2 3 3 3 4 4 4 1 5 5 6 6 6 7 7 7 8 8 8 5 9 2 5 10 8 3 11 6 1 12 4 7

Figure 10-7. Points and Lines Data Structure for the Cube in Figure 10-6.

Hidden Line Removal Representations Removing hidden lines (curves) involves removing lines and parts of lines that are obscured by faces of the object. The data structure must store more than just lines and points, it must know about faces. Therefore, we must add a face structure that represents a surface that is bounded by edges. In the cube of Figure 10-6, the faces are planes bounded by lines, called polygons. Polygons are the simplest form of a face and are very common among 3D data representations. A first storage method for polygons might be a list of points, ABCD, ADHE, EFGH, BCGF, ABFE, CDHG, where each adjacent pair of points in the list, including the last to the first, is a line. This suffers from the inefficiencies in space and redundant computations noted previously for lines. Alternatively, a polygon can be stored as a list of references to vertices, similar to the points and lines logic, called a points and polygons structure. The lines structure of Figure 10-7 is replaced by a polygons structure that has four indices per polygon (for this example) corresponding to the four vertices. Edges would be understood implicitly to be lines between adjacent vertices, requiring that the list of vertices be properly ordered.

Hidden Surface Removal Representations The data required for hidden surface removal is an extension of that for hidden line removal. In addition to the geometry of each face of the object, the light reflectance properties of the face also must be stored. Also, the lighting conditions of the surrounding environment are

Chapter 10. 3D Rendering

10.3 Planar Geometric Projections

135

needed. These will be discussed in later chapters.

10.3 Planar Geometric Projections

Planar geometric projection involves transforming 3D coordinates onto a projection plane (also called image plane and picture plane) in space and using the resulting plane coordinates to draw the image. There are two forms of planar projections, parallel and perspective: 1. Parallel Planar Projection Parallel projections have the center of projection, the eye point toward which points are projected, infinitely far from the projection plane. There are two kinds of parallel projections: A. Orthographic Parallel Planar Projection: The projection direction is normal to the projection plane. There are two kinds of orthographic projections: Axonometric: the projection plane is normal to a principal plane, i.e. top, front and side views. Isometric: the projection plane is not normal to a principal plane. B. Oblique Parallel Planar Projection: The projection direction is not normal to the projection plane. cabinet: cot(angle) = 1/2 (angle is about 63 degrees) cavalier: 30 or 45 degrees. 2. Perspective Planar Projection The center of projection is a finite distance from the projection plane.

10.4 Perspective Projection

First we will study wireframe perspective projection. Consider the eye coordinate system XYZe shown in Figure 10-8. Let the Ze=0 (XY) plane be the projection plane. A point in 3D space is projected onto the projection plane by finding the intersection with the plane of a line from the data point in question (X,Y,Z) to a given center of projection (COP). This line is called a projector. Thus (X1p,Y1p) is the projected 2D image of

Chapter 10. 3D Rendering

10.4 Perspective Projection

136

(X1,Y1,Z1) (X1p,Y1p) (X2p,Y2p) (X2,Y2,Z2) Xe

projectors

d center of projection COP at (0,0,d)

Figure 10-8. The Eye Coordinate System and Projected Lines. (X1,Y1,Z1), and similarly for (X2p,Y2p) and (X2,Y2,Z2). After perspective projection, lines remain lines and planes remain planes (this will be shown mathematically later). Therefore, the projection of a 3D line from point 1 to point 2 is the line connecting the projected endpoints. Angles, however, are not preserved after perspective projection. To derive the equations for perspective projection, first look down the Xe axis (Figure 109). The COP is a point on the Ze axis that is a distance d, called the perspective distance, from Ye (X,Y,Z)

COP d

Yp Xe (out of page)

Figure 10-9. Viewing a Projector Looking Down the Xe Axis. the projection plane. Using similar triangles, relate Yp to d, Y and Z: d Yp = Y ---------dz

Chapter 10. 3D Rendering

10.4 Perspective Projection

137

Now find Xp by looking down the Ye axis (Figure 10-10). (X,Y,Z) Ye (out of page) Xp Xe

COP Ze Figure 10-10. Viewing a Line Looking Down the Ye Axis. z 1 d Xp = X ---------- = X 1 -- d dz d 1 ---------- , also shown as ---------------- is the perspective factor and is a function of the perspective dz z 1 -- d distance d of the eye from the projection plane and the z coordinate of the given point. After perspective, when the data lies on the projection plane, we can use the standard twodimensional techniques to draw it. Examine the perspective factor more closely. 1. When z -, i.e. for data points far from the eye: d lim ---------- = 0 d z z Therefore (Xp,Yp) (0,0). The points converge to the origin. 2. When d , i.e. when observing from infinitely far away: d lim ---------- = 1 d z d Therefore (Xp,Yp) (X, Y). This is parallel (orthographic) projection. So, perspective projection from infinitely far away becomes parallel projection.

Chapter 10. 3D Rendering

10.4 Perspective Projection

138

When z > d, i.e., the data point is behind the COP: d ---------- = a negative number d z Therefore (Xp, Yp) = (a negative number) (X, Y). This causes inversion of the image, or

false projection. Ye

Figure 10-11. Inversion of Data Behind the COP. 4. When z d, i.e. a point in the z plane of the COP: d lim ---------- = d z zd The perspective factor approaches infinity, causing a severe lens distortion effect as objects approach the COP plane. Computationally, this is a problem that must be handled in code, as described later. 5. When z = 0, i.e. data points on the projection plane: d ---------- = 1 d z Points on the projection plane are not affected by perspective. Figure 10-12 summarizes these different regions along the Z axis and their projection characteristics.

Chapter 10. 3D Rendering

10.4 Perspective Projection

139

z>d

z=d

0<z<d

z<0 Ye

Figure 10-12. Effects of Data Z Values on Projection. There are other definitions for the eye coordinate system. One of the early definitions for the eye coordinate system places the COP at the origin and the projection plane at the location z=d in a left-handed coordinate system as shown in Figure 10-13. This is popular because Ze increases Ye Ze (into page) projection plane @ z = d Xe COP Figure 10-13. A Left-Handed Eye Coordinate System. away from the eye, which some consider better than a right-handed system where Ze decreases away from the eye. To derive the perspective factor for this system, again look down the Xe axis (Figure 1014). Ye (X,Y,Z) Yp Y ------ = --d Z so: Ze d Y p = Y -Z

Yp Xe (out of page)

Figure 10-14. Looking Down Xe in the Left-Handed Eye Coordinate System.

Chapter 10. 3D Rendering

10.5 3D Modeling Transformations

140

d Similarly, we find Xp = X -- . This appears simpler than the previous perspective equations, but has Z limitations for some applications, such as hidden surface removal, as well see later.

10.5 3D Modeling Transformations

Consider 3D transformations, this time going directly to matrix and functional forms. The 3D vector consists of four elements: (x, y, z, 1), and the homogeneous transformation matrix is four by four. 1. Translation: T(Tx, Ty, Tz) 1 0 x' y' z' 1 = x y z 1 0 Tx 2. Scaling : S(Sx, Sy, Sz) Sx 0 x' y' z' 1 = x y z 1 0 0 3. Rotation: RX(), RY(), RZ(). There are three basic rotations about each principal axis: Rotation about X, RX(): 1 0 0 0 0 cos sin 0 x' y' z' 1 = x y z 1 0 sin cos 0 0 0 0 1 Rotation Matrix Clues: 1. 2. 3. Rotating about X (Y, Z) does not change X's (Y's , Z's) Row 4 and column 4 are always the same. The problem is to locate the cos (C) and sin (S) elements: C S 0 Sy 0 0 0 0 Sz 0 0 0 0 1 0 1 0 Ty 0 0 1 Tz 0 0 0 1

Chapter 10. 3D Rendering

10.6 Homogeneous Coordinates

141

and to place the minus sign before one of the S terms. To do this, rotate an obvious point, like (0,1, 0) about the axis by 90 . For X: Y (0,1,0) 90 X Z (0,0,1)

Figure 10-15. Test Rotation about the X Axis. Rotation about Y, RY(): Y cos 0 x' y' z' 1 = x y z 1 sin 0 0 sin 0 1 0 0 0 cos 0 0 0 1

(1,0,0) (0,0,1) Z 90

Rotation about Z, RZ(): Y cos sin 0 0 sin cos 0 0 x' y' z' 1 = x y z 1 0 0 1 0 0 0 0 1 Z It is possible to rotate about an arbitrary axis, but this requires the concatenation of several basic rotation transformations. (0,1,0) 90 X (1,0,0)

10.6 Homogeneous Coordinates

We have simply ignored the fourth coordinate of vectors and the last column of matrices.

Chapter 10. 3D Rendering

10.6 Homogeneous Coordinates

142

For representing translation in matrix form and to make the transformation matrices square, we have set the fourth coordinate of a vector to 1, (x,y,z,1), and the last column of matrices to (0,0,0,1)T. Now we generalize this notation and give deeper geometric interpretation to it. A point in 3D space is a projection of a point in 4D space given by the coordinates (wx, wy, wz, w), the homogeneous coordinates of a point. Define normalized homogeneous coordinates or ordinary coordinates as coordinates having a unity value of w. The x, y and z components of homogeneous coordinates are the 3D coordinates only when w=1 (or 0 as will be discussed later), i.e. only after the coordinates have been normalized. Thus, (2, 4, 6, 2), (0.5, 1.0, 1.5, 0.5) are the same normalized coordinates, (1, 2, 3, 1), which is the 3D point (1,2,3). To find the normalized 3D coordinates, simply divide by the w coordinate, i.e., wx wy wz ------ , ------ , ------, 1 - - w w w To see the benefits of this convention, examine a general 4 by 4 homogeneous transformation matrix (for row vectors): a d g l 1. 2. 3. (a-i): (l-n): s: b e h m c f i n p q r s

Scaling and rotation elements. Translation. This is the homogeneous scaling element. Examine what happens when we

transform a normalized coordinate by a matrix with a variable s element: 1 0 xyz1 0 0 0 1 0 0 0 0 1 0 0 0 xyzs 0 s

After normalization, the coordinates are x y z 1 . The net effect is to uniformly scale the -- -- - - s s s 3D coordinates by s-1.

Chapter 10. 3D Rendering

10.6 Homogeneous Coordinates

143

p,q,r: These are projection elements. Begin by examining the effects of a transformation with variable r and s elements: 1 0 xyz1 0 0 Now normalize, x y z x y z ( rz + s ) ------------ ------------ ------------ 1 , only if rz + s 0 rz + s rz + s rz + s 1 Letting s = 1 and r = -- , where d is the perspective distance, d 1 1 1 x ----------- , y ----------- , z ----------- , 1 z z z 1 -- 1 -- 1 -d d d This is perspective projection! Thus, due to the convention that normalization introduces 0 1 0 0 0 0 1 0 0 0 x y z ( rz + s ) r s

division, perspective can be represented as a transformation in matrix form. The perspective matrix, P, is: 1 0 0 0 0 1 0 0 P = 1 0 0 1 -d 0 0 0 1 Where is the point (x,y,z,0)? Following the homogeneous convention, normalize the point: x y z 0 -- -- -- -- - - 1 0 0 0 0 The interpretation of the homogeneous vector (x,y,z,0) is a point at infinity on a line from the origin through (x,y,z). In other words, the vector (x,y,z,0) is a direction vector. This is a consistent interpretation when one considers that subtraction of two position vectors with the same w coordinates will result in a direction vector with a 0 w coordinate. The ability to represent points at infinity is another benefit of homogeneous representation.

Chapter 10. 3D Rendering

10.7 Viewing Transformations

144

Project the point (0, 0, d, 1), the COP for perspective projection: 1 0 0 0 0 1 0 0 0 0 d1 1 = 0 0 d 0 0 0 1 -d 0 0 0 1 This is the point at infinity on a line from the origin through (0,0,d,1). Therefore, the COP projects to the point at infinity on the Ze axis.

10.7 Viewing Transformations

The projection of 3D data onto the projection plane z = 0 requires that the data be defined relative to the eye coordinate system, where the observers eye, the COP, is at a point (0,0,d) and is looking toward (0,0,0). In general, one may want to observe data from some other position and direction that, for example, can be specified as a view vector. The view vector is given most naturally in world coordinates, coinciding with the data reference. Consequently, data must be Yw View Vector EYE point AIM point Xw
Xi,Yi,Zi Data

Zw Figure 10-16. World Coordinate System and View Vector. transformed from world coordinates into eye coordinates before projection can be performed (analogous to translating before scaling when scaling-about-a-point ). This transformation, called the viewing transformation, V, relates the eye coordinate system to the world coordinate system. Where is the eye coordinate system? One answer is to use the view vector to define the eye coordinate system. We will stipulate that AIM is the origin of XYZe and EYE is the COP, or

Chapter 10. 3D Rendering

10.7 Viewing Transformations

145

AIMw (0, 0, 0)e EYEw (0, 0, d)e Yw Ye

Xe EYE AIM

Xw Data in World Coordinates Zw Figure 10-17. World and Eye Coordinate Systems. For now, also make the arbitrary specification that the Ye axis will be vertical, i.e. the plane Xe=0 will be perpendicular to the plane Yw=0. This means that the observer stands up-right with respect to Yw when viewing the data in the world coordinate system. The plane Ze=0 is the projection plane and the Ze direction is the normal to the projection plane. The definition of a viewing transformation, V, is that it transforms a point in world coordinates (Pw) into the corresponding point in eye coordinates (Pe): P e = Pw V To derive V given the view vector, visualize the process as transforming the view vector as a line in world coordinates. We want to align the Ze axis on the view vector with the Zw axis. This can be accomplished using three basic transformations: 1. 2. 3. Translate the vector so that AIM goes to the origin, i.e. AIMw (0, 0, 0)e Rotate the result so that the translated EYE, EYE', swings into the Xw=0 plane. Rotate this result so that EYE'' swings down onto the Zw axis.

The matrix equation becomes: Pe = Pw T(?) R?(?) R?(?) So that, V = T(?) R?(?) R?(?)

Chapter 10. 3D Rendering

10.7 Viewing Transformations

146

Step one. This is a translation: P' = Pw T(-AIM) After this translation, the situation is as shown in Figure 10-18. Yw

EYE'=EYE-AIM EYE'' EYE'x Zw Figure 10-18. View Vector After Translation. Step 2. Rotate by angle about Yw to rotate EYE' into the Xw=0 plane. Using trig:
1 EYE' x = tan -------------- EYE' z

EYE'y EYE'z

Check the sign by noting the rotation direction assuming positive values. Note that EYE'z can be 0, and that the ATAN function returns angles between - -- and -- . Therefore we must code 2 2 this using the ATAN2(EYE'x,EYE'z) function. There is still a problem, however, in that EYE'x and EYE'z will both be zero when the view vector is parallel to the Yw axis. In this case, ATAN2 cannot be called. The solution is to set = 0. Step 3. Rotate EYE'' down by angle to (0,0,d) by rotating about Xw. EYE' y = tan --------------------------------------------- 2 2 EYE' x + EYE'z
1

To check the transformations, try transforming the data points AIM and EYE. You know what the results should be: (0,0,0) and (0,0,d). For example, consider Figure 10-19. We first compute EYE' = EYE - AIM = (0,1,0). Then,

Chapter 10. 3D Rendering

10.7 Viewing Transformations

147

Yw Ze View Vector EYE=(0,1,0) Ye Xe Xw Zw AIM=(0,0,0)

Figure 10-19. Example View Vector. following the steps above,

1 EYE' x 1 0 = tan -------------- = tan -- = 0 (by assignment) 0 EYE' z EYE' y 1 1 1 = tan --------------------------------------------- = tan --------------- = -- 0 + 0 2 2 2 EYE' x + EYE'z

V = T ( 0, 0, 0 )R Y ( 0 ) R X -- 2 1 0 0 0 0 1 0 0 0 0 0 1 1 0 0 cos sin 0 --2 2 V = ( I )( I) = 0 0 0 1 0 sin cos 0 --2 2 0 0 0 0 0 1 As a check, we transform AIM and EYE, AIM V = (0,0,0), EYE V = (0,0,1). Also note that the XYZe axes, treated like data, will align with the XYZw axes when transformed by V. Often you can do this by observation. The EYE-AIM view vector method just described can be far too restrictive for some applications. Imposing the condition that d is the distance between EYE and AIM can be awkward. It may be useful to consider this EYE-AIM view vector approach as an internal representation for the viewing information and give the user alternative methods for specifying the viewing position, direction and perspective distance. An example of one alternative is using polar (or spherical) coordinates: R, azimuth and

Chapter 10. 3D Rendering

10.7 Viewing Transformations

148

elevation, for a viewing sphere with a given center C. This is useful for walking around an object. We treat the data as though it were at the origin of a sphere of radius R. Consider viewing the earth as an example. Azimuth is the angle of rotation about the earths axis, i.e. east-west rotation, which in XYZw notation means a rotation about our Y axis. Elevation (also called inclination) is the angle from the equator, i.e. north-south rotation, that is a rotation about our X axis. Y

elevation (inclination) Yw center C Xw Zw Z azimuth X Radius R

Figure 10-20. Viewing Sphere and Polar Coordinates. The radius of the viewing sphere, R, and its center, C, are usually computed to surround the 3D data. One can compute a bounding sphere for the data, or more simply, compute a parallelepiped with corners (xmin,ymin,zmin) and (xmax,ymax,zmax). The sphere center C is then the midpoint of the diagonal connecting the corners, and R is half the length of the diagonal. Given the viewing sphere, the user specifies viewing information as elevation up, azimuth to the right. The problem is to compute EYE and AIM from this information, and then compute V as before. One approach is to set AIM = C, i.e. always look toward the center of the sphere. After all, this is the center of the data. To compute EYE, think of transforming a direction vector of length d along the Z axis (parallel to Zw) through C by the elevation and azimuth rotations, then adding this vector to AIM: EYE = (0,0,d,0) RX( -elevation ) RY( azimuth ) AIM

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

149

10.8 Three Dimensional Clipping

There are several approaches to three-dimensional clipping, all of which have as their objective clipping lines in 2D and/or 3D to remove parts of lines that would fall outside the window. Three approaches will be presented in the following sections. The approaches progress from a simple extension to 2D clipping, which serves as an introduction to 3D clipping, to the final objective of full 3D clipping in homogeneous coordinates.

10.8.1 Z-Clipping with 2D Window Clipping

We saw that as z d, perspective causes severe distortion, i.e. d ---------- dz and z > d produced an inverted image. Also, data far from the observer (ze << 0) become possibly distracting dots at the origin. To eliminate these effects, two clipping planes parallel to the projection plane are introduced: 1. 2. a near clipping plane at z = znto eliminate inversion and distortion, a far clipping plane at z = zf to provide a limited sight distance.

Clipping on the near plane is required prior to projection to avoid infinite projection factors and inverted data. Clipping on the far plane is optional and is useful for eliminating data too far to see and to produce special effects such as cross-sectioning. Clipping on the near and far planes is termed z-clipping. One approach to 3D clipping, based on the idea of utilizing existing 2D software for clipping and window to viewport mapping, is to: 1. clip the 3D line on the near and far planes (z-clip), 2. project the line into 2D on the projection plane, 3. clip the projected line against the 2D window, 4. and then map it to the viewport. Z-clipping can be implemented using logic based on the 2D Cohen-Sutherland clipping algorithm. The endpoint classification can be a simple Boolean value that signifies visible or invisible. This code is computed based on the z coordinate alone as illustrated in Figure 10-21.

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

150

near plane invisible visible 1 z > zn 1' 2 Ye

far plane invisible 2 1' z < zf 1

Ze zn Xe zf

Figure 10-21. Z-Clipping Endpoint Regions. To compute a clipped endpoint, use the symmetric equations for a 3D line: y y1 z z1 x x1 ---------------- = ---------------- = --------------x2 x1 y2 y1 z2 z1 where (x,y,z) are the coordinates of any point on the line between the known points (x1,y1,z1) and (x2,y2,z2). For a line crossing one of the planes, we know that z', the clipped z coordinate, will be zn on the near plane and zf on the far plane. Therefore, the x' and y' of point 1' shown in Figure 10-21 can be computed. Lines can be rejected as wholly invisible during z-clipping and thus would not be further processed. The process of drawing a line resembles a pipeline of operations starting with the endpoints in 3D world coordinates and ending with a clipped, projected line in 2D viewport coordinates. This is often called the 3D pipeline. For the Z-Clipping approach to 3D clipping, the pipeline appears as shown in Figure 10-22. Across the top of the pipeline element boxes are parameters that control the execution of each element of the pipeline. Across the bottom are the dimensionality and reference frame for the coordinates entering and exiting each element. The diagram in Figure 10-22 can be used as a model for coding the pipeline. Each block can be coded as a routine that inputs the endpoints of the line in 2D or 3D. The first routine, e.g. Draw3DLine( x1, y1, z1, x2, y2, z2 ), would be called to draw a line given the 3D local

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

151

modeling transformations M

view vector

zn & zf

window (wcx,wcy, wsx,wsy)

viewport (vcx,vcy, vsx,vsy)

CTM

Z-CLIP

2D CLIP

W-V MAP

3D local

3D world

3D eye

3D eye invisible (exit)

2D 2D projection window plane invisible (exit)

2D device

Figure 10-22. The 3D Pipeline for Z-Clipping. coordinates of the endpoints. The application can then use this routine without concern for all the downstream details of its operation, thus forming a 3D package of sorts.

10.8.2 3D Frustum Clipping

Rather than perform clipping in two steps, 3D Z-clipping and then 2D window clipping after projection, the process can be accomplished in one step. This consolidates the elements of the pipeline for both efficiency and simplification of coding, but requires somewhat more complex computations. Begin by visualizing four planes, called top, bottom, left and right, that pass through the COP and two corners of the window on the projection plane as shown in Figure 10-23. These four planes form a viewing volume, a semi-infinite four-sided pyramid extending from the COP in the viewing direction. When the viewing volume is combined with the near and far planes it forms a six-sided viewing frustum bounded by the top, bottom, left, right, near, and far clipping planes. Points inside the viewing volume will be visible and points outside it will be invisible. Therefore, clipping a line against the 6 sides of the viewing frustum prior to projection will guarantee that the final 2D line lies within the window after projection. Imagine holding the viewing volume up to your eye and aiming it somewhere. The side clipping planes limit your field-of-view in the horizontal and vertical projection plane directions.

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

152

Top Clipping Plane

Left Clipping Plane Window (z=0)

Far Clipping Plane (z=zf)

Xe Right Clipping Plane Ze COP Bottom Clipping Plane Near Clipping Plane (z=zn) Figure 10-23. Viewing Frustum. The horizontal Field-Of-View angle (FOV) is the angle between the left and right clipping planes (see Figure 10-24) and the vertical FOV is the angle between the top and bottom clipping planes. The FOV angle is often used as a viewing parameter.

Far Ye Left Right Xe

Field-Of-Fiew (FOV) angle Near COP Ze Frustum viewed down Ye

Figure 10-24. Horizontal Field-of-View Angle. Examine the top and bottom clipping planes of the viewing frustum when viewed down the Xe axis as shown in Figure 10-25. The equations of the top and bottom planes are functions of z, the window size wsy, and the perspective distance d. These are infinite planes that divide 3D space in the Y direction into 3 regions: above the top clipping plane, visible, and below the bottom

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

153

clipping plane. TOP clipping plane z y = wsy 1 -- d Ze wsy Xe (out) zn wsy zf z y = wsy 1 -- d invisible BOTTOM clipping plane

invisible

d COP

Figure 10-25. Side View (Looking Down Xe) of the Viewing Frustum. Similarly, Figure 10-26 shows a top view of the viewing frustum and the equations for the left and right clipping planes. These are infinite planes that divide 3D space into three regions, left, visible, and right. LEFT clipping plane zf wsx z x = wsx 1 -- d invisible COP Ze Figure 10-26. Top View (Looking Down Ye) of the Viewing Frustum. z There is a common factor, (1 - -- ), that should look familiar. It is the inverse of the d perspective factor. The 6 infinite planes form the boundaries of a visible volume that is described by the following equations. wsx
Ye (out)

RIGHT clipping plane

Xe z x = wsx 1 -- d invisible

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

154

z z Letting wsxp = wsx 1 -- , wsyp = wsy 1 -- : d d -wsxp -wsyp zf <= <= <= x y z <= <= <= wsxp wsyp zn

These equations can be coded into a 3D endpoint code using a 6-bit binary code, where each bit represents the Boolean term the point lies on the invisible side of the ( TOP, BOTTOM, LEFT, RIGHT,NEAR, FAR) clipping plane. A zero code means the point is inside the frustum. 32 NEAR z > zn 16 FAR z < zf 8 4 2 1

TOP BOTTOM LEFT RIGHT y > wsyp y < -wsyp x < -wsxp x > wsxp

Figure 10-27. Sample 3D Endpoint Region Code. The frustum clipping algorithm can be implemented as an extension of the CohenSutherland 2D clipping algorithm. The exit conditions are still based on the AND of the endpoint codes. Clipping a line that crosses a boundary is a bit more complicated in 3D. Given two endpoints, P1 and P2, where P1 is invisible and P2 is visible, i.e. the line P1 P2 crosses a given clipping plane, one must compute a new P1 that lies on this plane. This was done in the 2D clipping algorithm using the symmetric equations of a line. Unfortunately, we now have a line in 3D space that intersects an oblique plane (not orthogonal to a principal axis). One technique for this computation is to use the plane equation of the clipping plane, Ax + By + Cz + D = 0. First note that this equation can be expressed as a homogeneous dot product: A B = 0 xyz1 C D

or,

PB = 0

B is called the plane coefficient vector for a clipping plane. The equation of any point along the line P1 P2 can be expressed as a parametric vector function P(), where the scalar parameter varies between 0.0 at P1 to 1.0 at P2:

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

155

P ( ) = P1 + ( P2 P1 ) First form the homogeneous dot product of the P() equation with the plane coefficient vector of the violated clipping plane, B. Then solve for the value of the parameter for which the line intersects the plane by recognizing that P() B = 0 at the intersection (i.e., when P() is on the plane). P() B = 0 P1 B + ( P2 P1 ) B = 0 P1 B = -----------------------------------P1 B P2 B The clipping plane coefficient vectors, B, are formed from the equations of the clipping planes and are shown in Figure 10-28. Plane Coefficient Vector ( 1, 0, ( ------- ) , wsx ) ( -1, 0, ( ------- ) , wsx ) ( 0, 1, ( ------- ) , wsy ) ( 0, -1, ( ------- ) , wsy ) ( 0, 0, 1, -zf ) ( 0, 0, -1, zn )
wsy d wsy d wsx d wsx d

Plane Identifier LEFT RIGHT BOTTOM TOP FAR NEAR

Algebraic Equation
z x = wsx 1 -- d z x = wsx 1 -- d z y = wsy 1 -- d z y = wsy 1 -- d

Endpoint Test (Invisible is True)

z x < wsx 1 -- d z x > wsx 1 -- d z y < wsy 1 -- d z y > wsy 1 -- d

z = zf z = zn

z < zf z > zn

Figure 10-28. Frustum Clipping Plane Coefficient Vectors. Using the fact that the plane equation, Ax+By+Cz+D=0, is the same as its negation, -AxBy-Cz-D=0, the coefficient vectors can be formed so that the value of the dot product, P B, can be used to determine on which the side of the plane the point P lies. The plane coefficient vectors shown in Figure 10-28 have been formed such that (P B) yields a positive number (or zero) if P is on the visible side of the plane and a negative value if the point is on the invisible side. As a result, the endpoint code test also can be based on the plane coefficient vector. This can further

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

156

simplify and modularize the frustum clipping code. A single general plane clipping routine can be passed the plane coefficient vectors of the six clipping planes, classify the endpoints and clip to the boundaries using only the plane coefficient vector. The pipeline for 3D frustum clipping is shown in Figure 10-29. view vector zn, zf , d & window d viewport

CTM

Frustum Clip

W-V Map

3D local

3D world

3D eye

2D window

2D device

Figure 10-29. 3D Pipeline with Frustum Clipping.

10.8.3 3D Clipping in Homogeneous Coordinates

A significant disadvantage of 3D frustum clipping is that the plane coefficient vectors vary with the viewing parameters (zn, zf, wsx,wsy,d) and must be re-computed whenever this viewing information changes. Also, note that except for clipping, the pipeline consists of transformations that could be concatenated into one matrix. The entire pipeline could become a vector to matrix multiplication followed by a simple normalization (perspective division) of a homogeneous vector. The problem, however, is clipping. Side plane clipping is not the real problem, this can be done after projection. The real problem is z-clipping on the near clipping plane that must be done before perspective to avoid inversion and division by zero. Looking more closely at this problem, perspective projection of a point occurs when the homogeneous vector is normalized to compute the 3D coordinates, i.e., when wx, wy and wz are divided by the w coordinate. Therefore, any 3D clipping computations done after transformation by the perspective matrix P must be performed prior to this normalization, when the coordinates are in homogeneous 4D space, i.e. (wx,wy,wz,w) with w not necessarily equal to one. This is not

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

157

as complex as it first sounds, although it may appear to be more complex than the previous methods due to the nature of 4D coordinates. Observe what happens to the viewing frustum after perspective projection (Figure 10-30). Ye Yp

P window (wsx,wsy,0) Xe z=zf

Xp
z zf ( d ) = ----------fp d zf

Ze eye

z=zn

Zp eye at
z

z (d) n = -----------np dz n

viewing frustum before perspective

viewing frustum after perspective

Figure 10-30. Viewing Frustum Before and After Perspective. We know that the window is not affected by projection (because z=0) and the COP goes to z=. Therefore the frustum transforms into a right parallelepiped, with the size clipping planes extending through the window between the projected near and far planes. Further examining the process, the two-dimensional image of the data is actually an orthographic parallel projection (front view, x and y values only) of the projected 3D coordinates. Deeper insight into the perspective projection process is gained by examining three regions along the Z axis in the W-Z plane of 4D space before and after perspective, as shown in Figure 1031. Before projection, the coordinates of point A are (-,-,d,1), where - means any value. The coordinates of B are (-,-,0,1). After projection, A is (-,-,d,0), a point at infinity. B is (-,-,0,1), unchanged as expected because it is on the projection plane. Region 1. 2. 3. Before Perspective d<z< 0<z<d - < z < 0 After Perspective - < z < -d 0 < z < + -d < z < 0

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

158

Before Projection W Z z=d COP z= -d 1 2 w=1 plane 3 COP Z 2

After Projection W 3 z=d

old w=1 plane w=1 1 z = -d -

Figure 10-31. W-Z Plane View of Perspective Projection. Several of the z projections involve limits. In region 1, as z approaches d, zd lim ---------- = d z zd
z>d

Similarly, as z approaches and - in regions 1 and 3, zd lim ---------- = d d z z Finally, as z approaches d in region 2, zd lim ---------- = d z zd
z<d

zd lim ---------- = d d z z

In region 1, points behind the COP prior to projection transform into points in front of the eye after projection. Points in region 2 between the COP and the projection plane are projected into greatly expanded semi-infinite region. Consider what happens to a point that is placed very near the COP. It will project to a point near the eye, i.e., near . Likewise, points in the semi-infinite region 3 are compressed into a region between 0 and -d. This shows the highly non-linear nature of perspective projection. Homogeneous clipping can be simplified considerably by transforming the coordinates after applying perspective using a normalization transform, N, that transforms a point into clipping coordinates. The x, y, and z coordinates of a point transformed by P are mapped into a clipping coordinate system, XYZc, in which the visible coordinates, coordinates that were originally inside

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

159

the viewing frustum in XYZe, lie inside a cube centered at the origin (x,y,z [-1,1]). In effect, the view volume is transformed into a cube, as shown in Figure 10-32. After clipping lines (in 4D) against the clipping cube, they can be projected and mapped directly to the viewport. Yp far clipping plane window N
(1,1,1)

Yc far clipping plane

Xp (-wsx,-wsy, zfp) Zp eye at (wsx,wsy, znp) near clipping plane

(-1,-1,-1)

Zc near clipping plane eye at the clipping cube

viewing frustum after perspective

Figure 10-32. Homogeneous Normalization After Perspective. The normalization transformation N is constructed in two steps: (1) translate the center of the projected view volume to the origin, and (2) scale the resulting volume into a cube with x, y and z in [-1,1]. Referring to Figure 10-32, the center of the projected frustum lies along the z axis half-way between the near and far clipping planes. The locations of the near and far clipping planes usually are given as distances from the COP rather than z coordinates in XYZe, which would be awkward for a user. The distance to the near plane is n and the distance to the far plane is f. Thus, the z coordinates of the near and far clipping planes are zn = d-n and zf = d-f. Let znp be the z coordinate of the near clipping plane after projection, and zfp be the z coordinate of the far clipping plane after projection. Using the standard perspective equations: d z np = z n ------------d zn and d z fp = z f -----------d zf

Using the distances n and f in place of the z coordinates, d(d n) d(d f) and z fp = -----------------z np = ------------------n f

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

160

The half-sizes of the projected volume are wsx, wsy, and (znp-zfp)/2. We can now build N: ( z np + z fp ) 1 1 2 N = T 0, 0, -------------------------- S -------- , -------- , ------------------------- wsx wsy ( z z - 2 )
np fp

Concatenate the P and N matrices and multiply by d to simplify the terms. Notice that multiplying the matrix by d results in multiplying the resulting coordinates by d. However, scaling a homogeneous coordinate does not affect the normalized value. The resulting matrix, called dPN, is shown below. d -------- 0 wsx d 0 -------wsy dPN = 0 0 0 0 0 0 0

f+n 1 ---------fn df + dn 2fn 0 -------------------------------- d nf

d d Consider the scaling elements -------- and -------- above. Referring to Figure 10-24, we see that wsx wsy d FOVx -------- = cot -------------- 2 - wsx and d FOVy -------- = cot -------------- 2 - wsy

The perspective and normalization transformations transform eye coordinates into clipping coordinates, Pc = Pe dPN After transformation by dPN, the resulting homogeneous coordinates must be properly clipped against the clipping cube. Each endpoint is a homogeneous coordinate (wx, wy, wz, w) for which the visible 3D (projected) coordinates lie between -1.0 and 1.0. The ordinary 3D clipping volume is described by the equations: -1 <= wx wy wz ------ , ------ , -----w w w <= 1

Recall, however, that clipping must occur prior to division. Therefore, clipping must be performed in the 4-dimensional homogeneous space. The above equations in 4D space are: -w <= wx, wy, wz <= w

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

161

This involves clipping on the six planes wx = -w, wx = +w, and so on for wy and wz. There is one special action necessary for lines with one endpoint that has a negative w value. Note that if both endpoints have negative w values, the line is wholly rejected. Recall that a negative w value results from perspective transformation of a point with a z coordinate behind the z COP, i.e. w = 1 -- is negative when z > d. This line must be clipped against the w=0 plane, d removing the portion of the line behind the COP. Clipping on the w=0 plane must be done first, before clipping on the other planes. This is a simple matter of properly ordering the plane tests in the clipping code. Homogeneous clipping involves seven hyperplanes in 4D space. Once again, the clipping algorithm can be an extension of the basic Cohen-Sutherland 2D clipping logic, now clipping on a volume bounded by seven planes. A seven bit clipping code can be used, where the endpoint codes identify mutually exclusive regions bounded by the infinite planes called TOP, BOTTOM, LEFT, RIGHT, NEAR, FAR, WZERO (for the w=0 hyperplane). After checking the Boolean AND (quick reject) and OR (accept) of the endpoint codes, the line must be clipped by recomputing one of its endpoints. Computing the intersection of the line with a clipping plane follows the same logic as described previously for frustum clipping. The plane equations, however, are much simpler and contain only constants (by design). Be sure to use homogeneous coordinates now. A B = 0 wx wy wz w C D

or,

P Bi = 0

The clipping plane coefficient vectors for the homogenous clipping coordinates, formed as before so that P B <0 for invisible points, are given in Figure 10-33. Note that they are simple constants. The pipeline for 3D homogeneous clipping consists of two matrix multiplications and the homogeneous clipping procedure: The homogenous line clipping algorithm is outlined in Figure 10-35.

Chapter 10. 3D Rendering

10.8 Three Dimensional Clipping

162

Plane Identifier WZERO LEFT RIGHT BOTTOM TOP FAR NEAR

Plane Coefficient Vector ( 0, 0, 0, 1) ( 1, 0, 0, 1) ( -1, 0, 0, 1) ( 0, 1, 0, 1) ( 0, -1, 0, 1) ( 0, 0, 1, 1) ( 0, 0, -1, 1)

Algebraic Equation w=0 wx = - w wx = w wy = - w wy = w wz = - w wz = w

Endpoint Test w<0 wx < - w wx > w wy < - w wy > w wz < - w wz > w

Figure 10-33. Homogeneous Clipping Plane Data. M1 Homog. Clip & Divide M2 dPN Viewport Map

CTM

(x,y,z,w)local

(x,y,z,w)clip

(x,y)device

Figure 10-34. 3D Pipeline with Homogeneous Clipping.

Chapter 10. 3D Rendering

10.9 EYE-AIM Viewing Methods

163

10.11 Review Questions

168

10.11 Review Questions

1. A special eye coordinate system places the COP at the origin and projection plane at z= -d in a right-handed eye coordinate system, shown below. Derive the perspective projection Ye

projection plane @ z = AIM COP Xe

Ze equations for this eye coordinate system. Show the P matrix. 2. Derive the perspective projection equations for x, y and z for a left-handed eye coordinate system with the center of projection at the origin and projection plane at z = +d. Derive the corresponding P matrix. 3. The computation of the eye position given an AIM position using polar coordinates was given earlier as EYE = (0,0,d,0) R(-elevation, X) R(azimuth, Y) AIM. Rewrite this a Pw = Pe M. Derive the viewing transform directly from the elevation and azimuth angles, without first computing EYE. 4. Why is it necessary to do near plane z clipping? Where in the 3D pipeline should it be done? Why is it desirable to do BACK plane z clipping? Where in the 3D pipeline should it be done? 5. The homogeneous clipping coordinates of the endpoint of a line after transformation by V, P and N are (1,3,-3,2). Identify and give the plane equation of each clipping plane that is violated by this point.

Chapter 10. 3D Rendering

10.11 Review Questions

169

Develop the perspective P and normalization N matrices for transforming data from a righthanded eye coordinate system, with the projection plane at z=0, center of projection at (0,0,d), square window half-size of S, and near and back clipping plane locations ZF and ZB, into left-handed homogeneous clipping coordinates in which the visible x, y and z coordinates lie between 0.0 and 1.0 and z increases away from the eye. Draw and label an illustration of the 3D viewing frustrum before and after the P and N transformations. Show the N transformation as a concatenation of basic transforms given in functional form with symbolic expressions as arguments.

Chapter 10. 3D Rendering

11.1 Review of Vector Geometry Fundamentals

170

Chapter 11. Computational Geometry

It should be clear by now that geometry is at the root of virtually all computer graphics problems. Geometric modeling, the representation of physical objects in the computer, is also at the heart of two areas receiving considerable attention today: computer-aided design (CAD) and computer-aided manufacturing (CAM). CAD, CAM and computer graphics share a mutual interest in geometry. The term computational geometry describes an area of study addressing mathematical algorithms and computational methods for representing and manipulating geometry in the computer. We begin with a review of vector geometry fundamentals to establish terminology and basic methods on which later sections will build.

11.1 Review of Vector Geometry Fundamentals

There are significant advantages to using vector representation and vector equations to solve geometry problems instead of algebraic equations operating separately on each x, y and z coordinate: 1. 2. 3. 4. 5. more general solutions, 2D and 3D solution methods are the same, avoids many special cases, clearer geometric interpretation, code more closely resembles the mathematical form.

In Figure 11-1, the vector V is a direction vector that represents direction and length. It is also called a relative vector. The coordinates of a direction vector are relative displacements. The vector P in Figure 11-1 is a position vector, a vector that begins at the origin O. A position vector represents a location in space with respect to some coordinate system. The origin itself is a position vector. The distinction between direction and position vectors is sometimes subtle and often overlooked. It will be a useful distinction later when computational algorithms are discussed. We will represent all vectors in homogeneous coordinates: V = [ wx, wy, wz, w ].

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

171

V P O X

Z Figure 11-1. Position Vector P and Direction Vector V. Recall that the homogeneous vector [ x, y, z, 0 ] was reasoned to be a point at infinity on a line through the origin and the point. Such a vector also results from the vector subtraction of two position vectors with identical w coordinates: x2 y2 z2 w x1 y1 z1 w = x 2 x1 y 2 y1 z2 z1 0

Therefore, a direction vector has a zero w coordinate. As expected, position vectors have non-zero w coordinates. This will be particularly useful when we consider transformations, such as translation, applied to direction and position vectors. Geometrically, one can think of the vector coordinates as magnitudes to be applied to unit vectors along the principal axes in the vector equation (the dashed lines in Figure 11-1): V=xi+yj+zk where i, j, k are unit vectors, i.e. their length is one.

11.2 Basic Vector Utilities

In C and C++, one can define a type called vector that is an array, typedef double vector[4]; vector v; v[0] = 1.0; /* or v[X] = 1.0; */

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

172

or a structure: typedef struct { float (or double) x,y,z,w; } vector; vector v; v.x = 1.0; The structure vector form allows assignment of vectors, pass-by-value function arguments and vector return values from functions. The following sections develop a basic set of vector utilities. The use of the vector structure is exploited in the example functions. The functions pass vectors by value, which encourages clear and correct coding, but is inefficient compared to passing vectors by address.

11.2.1 Vector Magnitude

The magnitude or length of a vector is shown as |V| and computed as: V = x +y +z
2 2 2

double VMag( vector v ) { return sqrt(v.x * v.x + v.y * v.y + v.z * v.z); }

Figure 11-2. Vector Magnitude Function. Note that the magnitude function should be applied only to the direction vectors (w=0). The w coordinate is ignored. In our applications, the magnitude is meaningful only for ordinary 3D coordinates.

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

173

11.2.2 Vector Addition and Subtraction

Vector addition is shown as: V1 V2 = [ x1 + x2 , y1 + y2, z1 + z2, w1 + w2 ], and similarly for subtraction.

vector VAdd(vector v1, vector v2) { v1.x += v2.x; /* remember that v1 is passed by value */ v1.y += v2.y; v1.z += v2.z; v1.w += v2.w; return v1; } vector VSub(vector v1, vector v2) { v1.x -= v2.x; v1.y -= v2.y; v1.z -= v2.z; v1.w -= v2.w; return v1; }

Figure 11-3. Vector Addition and Subtraction Functions.

11.2.3 Vector Multiplication and Division by a Scalar

A vector is scaled by the factor f as : f V = [ f x, f y, f z , f w ].
vector VScale( vector v, double f ) { v.x *= f; v.y *= f; v.z *= f; v.w *= f; return v; }

Figure 11-4. Vector Scaling Function.

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

174

11.2.4 Unit Vector

To make a unit vector, a vector of length one, of a vector: V V = -----V
vector VUnit( vector v ) { v = VScale( v, 1.0 / VMag(v) ); v.w = 0.0; /* make a direction vector */ return v; }

Figure 11-5. Unit Vector Function. Here one sees the advantages of vector-based programming: compactness and the code resembles the mathematical expression. The unit vector operation should be applied only to 3D vectors, and sets the result to a direction vector (w = 0). Its application is to create unit magnitude vectors of direction vectors.

11.2.5 Dot Product:

The dot product, also called the scalar product, is computed as: V1 V2 x1 x2 + y1 y2 + z1 z2+ w1 w2.
double VDot( vector v1, vector v2 ) { return v1.x*v2.x + v1.y*v2.y + v1.z*v2.z + v1.w*v2.w; }

Figure 11-6. Dot Product Function. For direction vectors, the value of the dot product can be interpreted as (Figure 11-7): V1 V2 = | V1 | | V2 | cos , and can be used to obtain the cosine of the angle between vectors: ( x 1 x 2 + y 1 y2 + z1 z2 ) cos ( ) = --------------------------------------------------V1 V2 The dot product is commutative and distributive: V1 V2 = V2 V1

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

175

V | 2 |V

s co

V1 V2 = V2 cos | V1 |

Figure 11-7. Geometric Interpretation of the Dot Product. V1 ( V2 V3 ) = V1 V2 + V1 V3 The dot product also can be interpreted geometrically as the projection of one vector on another, as illustrated in Figure 11-7. This makes the dot product useful for some geometric reasoning computations. For example, if the dot product of two vectors is negative then cos() is negative because vector magnitudes are always positive, which means that must be between 90 and 270 degrees, so we can say that the vectors are facing in opposite directions. Similarly, if the dot product is zero, then cos() is zero, so = 90 or 270 degrees, so the vectors are perpendicular. Note that V V = | V | | V | cos(0) = | V |2.

11.2.6 Cross Product

The cross product, also called the vector product, is computed as a vector: ( y 1 z2 y2 z1 ) i j k x 1 y1 z1 ( x z x z ) 1 2 2 1 x 2 y2 z2 ( x 1 y 2 x2 y 1 )

V1 V2 =

The direction can be determined by observation using the right hand rule, i.e. close your right hand in the direction of the first to second vector and your thumb points in the direction of the cross product. The magnitude of the vector product is interpreted as: | V1 V2 | = | V1 | | V2 | sin . The cross product function should only be applied to direction vectors. The w coordinates are

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

176

V1 V1 x V2 Figure 11-8. Direction of Cross Product Vector.

vector VCross(vector v1, vector v2) { vector vout; vout.x = v1.y * v2.z - v2.y * v1.z; vout.y = v2.x * v1.z - v1.x * v2.z; vout.z = v1.x * v2.y - v2.x * v1.y; vout.w = 0.0; return vout; }

Figure 11-9. Cross Product Function. ignored. The cross product is not commutative, V1 V2 = - V2 V1 (opposite directions), but is distributive, V1 ( V2 V3 ) = V1 V2 V1 V3 . The cross product also is useful for geometric reasoning. For example, if the cross product of two vectors is a zero length vector, then (assuming the vectors magnitudes are non-zero) sin() must be zero, meaning =0 or 180 degrees. We can reason that the vectors are parallel and facing the same or opposite direction.

11.2.7 Vector Area

The area function is defined as the magnitude of the cross product of two vectors: Area(V1 ,V2 ) | V1 V2 |. This is best explained geometrically. Consider two direction vectors placed at a common point, as in Figure 11-10 We extend the vectors to form a parallelogram with V1 as the base. Define height as the projector from V2 perpendicular to the base. From the definition of the cross product

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

177

height base V1

Figure 11-10. Geometric Interpretation of the Area Function. magnitude, | V1 V2 | = | V1 | | V2 | sin(), but | V2 | sin() is the value of height, and | V1 | is the value of base. Therefore, the magnitude of the cross product is the area of the parallelogram, and hence the reason for the name Area. What is the geometric significance when the Area(V1,V2) = 0? By definition, the area of the enclosing parallelogram is zero, hence the vectors must be colinear (pointing in the same or opposite directions).
double VArea( vector v1, vector v2 ) { return VMag( VCross( v1, v2 ) ); }

Figure 11-11. The Vector Area Function.

11.2.8 Vector Volume

The vector volume function is another vector computation that has a useful geometric interpretation. Given three direction vectors, the definition of the vector volume function is the dot product of the cross product: Vol(V1 ,V2 , V3) ( V1 V2 ) V3. This is also known as the scalar triple product. The result, a scalar, is the volume of the parallelepiped formed by the extensions of the vectors. If Vol(V1 ,V2 , V3) = 0, then either two of vectors are parallel or the three vectors lie in a plane, i.e., contain no volume.

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

178

V2 V1 Figure 11-12. Vector Volume Function.

double Vol( vector v1, vector v2, vector v3 ) { return VDot( VCross( v1, v2 ), v3 ); }

Figure 11-13. Vector Volume Function.

11.2.9 Vector Equation of a Line

We form the vector equation of a point set that describes a straight line through point P in the direction V: P(t) = P t V, where t is a parameter (Figure 11-14). P Y Z X V

Figure 11-14. Vector Line Notation. Depending on the range of the parameter t used in the equation, we can classify P(t) as a: line ray segment if t can have any value, - <= t <=, if t is limited to t >= 0, if t is limited to t0 <= t <= t1 (usually 0 <= t <= 1).

11.2.10 Vector Equation of a Plane

We form the vector equation of a point set that describes a plane through point P0 with

Chapter 11. Computational Geometry

11.2 Basic Vector Utilities

179

normal direction N by observing that any point P in the plane forms a line in the plane with P0 that is perpendicular to the normal. The point-normal vector equation of the plane is given as: (P - P0) N = 0. N P0 P

Figure 11-15. Vector Plane Notation. Letting the plane normal N = (A, B, C), the plane point P0 = (x0, y0, z0) and an arbitrary point on the plane P = (x, y, z), look closer at this equation. First expand the dot product, (P N) - (P0 N) = 0. Now expand the vector equation into algebraic form, (A x + B y + C z) - (A x0 + B y0 + C z0) = 0. Compare this equation to the standard algebraic plane equation A x + B y + C z + D = 0. One sees that: D = - (P0 N) = - (A x0 + B y0 + C z0)

Chapter 11. Computational Geometry

11.3 Geometric Operations on Points, Lines and Planes

180

11.3 Geometric Operations on Points, Lines and Planes

Now apply the basic vector functions to solve a number of common geometric operations on points, lines and planes.

11.3.1 Distance from a Point to a Line

Given a line defined by the point P and direction vector V, and a point Q as shown in Figure 11-16, one can compute the distance to the line using the vector functions. Derive the computations using geometric interpretation and insight. As we did with the vector Area function, begin by

Q d P* P

Figure 11-16. Distance d from a Point Q to a Line P-V. forming a parallelogram with P, Q and V by duplicating the line PQ and the vector V to form four sides. The distance d is its height. The area of this parallelogram is the base times height, or |V| * d. Therefore, V (Q P) VArea ( V, VSub(Q, P) ) d = -------------------------------- = --------------------------------------------------------V VMag(V) Let P* = P t V, for some t. Then, P P (Q P) V VDot(VSub(Q, P), V) t = --------------------- = --------------------------- = ---------------------------------------------------V VV VDot(V, V) Examine the cases shown in Figure 11-17.

Chapter 11. Computational Geometry

11.3 Geometric Operations on Points, Lines and Planes

181

Q V t>1 P P* P* Q t<0 P V

Figure 11-17. Examples of Distance of Point to Line.

11.3.2 Distance Between Two 3D Lines

The distance between two 3D lines, i.e. the smallest distance, occurs along a line perpendicular to both given lines. First, observe that two lines will either be parallel or not. If V1 and V2 are parallel, they lie in a plane, so one can think of the area between them. P2 P2 - P1 P1 d V1 V2

Figure 11-18. Distance Between Parallel Lines. Using logic similar to that of finding the distance from a point to a line: V1 ( P2 P1 ) VArea ( V 1, VSub(P 2, P 1) ) d = --------------------------------------- = ---------------------------------------------------------------V1 VMag(V 1) If V1 and V2 are not parallel, then they form a volume for which the distance d is the height: Vol ( V 1, V 2, VSub(P 2, P 1) ) ( V1 V2 ) ( P2 P1 ) d = --------------------------------------------------------- = -----------------------------------------------------------------V1 V2 VArea ( V 1, V 2 ) Note that the VArea result above will be zero if the lines are parallel. This can be used as a decision value in the algorithm to compute the distance between lines.

11.3.3 Intersection of Two Lines

Two 3D lines intersect when the distance between them is 0 (approximately, of course). To

Chapter 11. Computational Geometry

11.3 Geometric Operations on Points, Lines and Planes

182

compute the point of intersection, P*, using vectors, solve the vector equation expressing the statement the point on line 1 that is the same as the point on line 2 : P1 (t) = P2 (s). P1(t) P* V2 P1 P2(s) V1

Figure 11-19. Intersection of Two Lines. The problem is to find the parameter values for t and s that solve this equation. Substituting the vector expressions for the lines yields a vector equation in the variables s and t: P1 t V1 = P2 s V2 Begin solving this equation by removing the absolute positions P1 and P2: t V1 = ( P2 - P1 ) s V2 Now cross both sides with V2 to eliminate s: t ( V1 V2 ) = ( (P2 - P1) V2 ) s ( V2 V2 ) Of course, ( V2 V2 ) = 0 by definition, so t ( V1 V2 ) = ( ( P2 - P1 ) V2 ). Dot this equation with (V1 V2) to make a scalar equation in t, t ( V1 V2 ) ( V1 V2 ) = ( (P2 - P1) V2) ( V1 V2). Finally, [ ( P2 P1 ) V2 ] ( V1 V2 ) t = -------------------------------------------------------------------------( V1 V2 ) ( V1 V2 ) Similarly, solving for s: [ ( P2 P1 ) V1 ] ( V1 V2 ) s = -------------------------------------------------------------------------( V 1 V2 ) ( V 1 V 2 ) The denominator of these expressions is zero for (V1 V2) = 0, so this must be a degenerate condition. Geometrically, the vector V1 V2 is the common normal between the two vectors. If this is 0, then the vectors must be parallel or lie along the same line, and possibly also

Chapter 11. Computational Geometry

11.3 Geometric Operations on Points, Lines and Planes

183

be overlapping. These must be handled as special cases in code. How do we know if the intersection point is on the segment P1 (t)? This is the advantage of parametric notation. The parameter value is a measure of the position of a point along the line. If 0 <= t <= 1, the point is along the segment P1 to P1+V1. If t < 0, the point is before P1, and if t > 1, it is after P1+V1. Similar logic holds for s and the line between P2 and P2+V2. This is not as messy to code as it looks, see Figure 11-20.
Boolean LineLineIntersection( vector p1, vector v1, vector p2, vector v2, double *s, double *t ) { double numerator, ndotn; vector n, p21; n = VCross( v1, v2 ); ndotn = VDot( n, n ); if( VMag(n) < TOOSMALL ) return FALSE; p21 = VSub( p2, p1 ); numerator = VDot( VCross( p21, v2 ), n); *t = numerator / ndotn; numerator = VDot( VCross( p21, v1 ), n); *s = numerator / ndotn; return TRUE; }

Figure 11-20. Line-line Intersection Function. With s and t known, P* is computed using P1(t) or P2(s). Figure 11-21 is an example of the use of the LineLineIntersection routine.
double s, t; vector pstar; if( LineLineIntersection(p1,v1,p2,v2,&s, &t) == TRUE ){ pstar = VAdd( p1, VScale( v1, t ) ); printf(the lines intersect at (%f,%f,%f)\n, pstar.x, pstar.y, pstar.z ); } else { printf(the lines do not intersect\n ); }

Figure 11-21. Example of the Use of LineLineIntersection.

11.3.4 Intersection of a Line with a Plane

Another common problem in computational geometry is finding the point of intersection

Chapter 11. Computational Geometry

11.3 Geometric Operations on Points, Lines and Planes

184

between a line and a plane. As in the case of the line-line intersection, this problem can be N P0 P Figure 11-22. Line and Plane Intersection. expressed geometrically as, there is a point on the plane that is also on the line. In vector notation, this becomes ( P* - P0 ) N = 0 P*(t) = P t V Substituting P*, [ ( P t V ) - P0 ] N = 0 Expanding: t ( V N ) + ( P - P0 ) N = 0 Finally, ( P P0 ) N t = ----------------------------VN The possibility of a zero in the denominator should always gain our attention. In this case, as in most vector-based computations, it is a warning of a special geometric situation. If V N is (approximately) zero, the line is parallel to the plane, so there can be no intersection. Of course, the line could be in the plane. If the plane is given as a plane coefficient vector, B, the line-plane intersection is computed as: PB t = ------------VB any point on the plane any point on the line V

Chapter 11. Computational Geometry

11.4 Change of Basis in Vector-Matrix Form

185

11.4 Change of Basis in Vector-Matrix Form

Consider the coordinates of a position P in two different coordinate systems, XYZ1 and XYZ2, as shown in Figure 11-23. Y1 Y2 P2 P1 O2 Z1 [0,0,0,1] X1 P X2 Z2

Figure 11-23. Two Coordinate Systems. Given a vector in XYZ1, P1, we would like to find its coordinates with respect to XYZ2, P2, using a transformation M: P2 = P1 M. The problem is to find the transformation M, called the change of basis. It is reasonable to assume that we know the origin of the second coordinate system, O2, and the unit direction vectors of the X2, Y2 and Z2 axes, called I2, J2 and K2, all with respect to XYZ1. This information allows us to form four equations with known vectors for P1 and P2, as summarized in the following table: P1 (XYZ1 Coordinates): I2 J2 K2 O2 P2 (XYZ2 Coordinates): [1,0,0,0] [0,1,0,0] [0,0,1,0] [0,0,0,1]

This set of equations can be expressed in matrix form, 1 0 0 0 0 1 0 0 0 0 1 0 I2 0 0 = J2 M 0 K2 1 O2

Chapter 11. Computational Geometry

11.4 Change of Basis in Vector-Matrix Form

186

where M is the unknown 4 by 4 matrix. Solving for M, I2 M = J2 K2 O2 Therefore, the transformation of a vector from one coordinate system to another can be found by finding the inverse of the matrix formed by putting the three unit direction vectors (in homogeneous form) in the first three rows and the origin position vector in the final row, of the second system with respect to the first. The general matrix inverse is a computationally expensive operation, so it is desirable to further refine this process by factoring the matrix on the right hand side into a rotation matrix and a translation matrix, I2 J2 K2 O2 = I2 J2 K2 I I I I O2
1

where the Is represent the respective row of the identity matrix. This can be done because we know that the only transformations in M are rotations and translations. The inverse of a product of matrices is the reverse order product of the respective inverses,
1 1 1

I2 M = J2 K2 O2 =

I I I O2

I2 J2 K2 I

and the inverse of a translation is simply the translation by the negative vector. The rotation matrix is an orthonormal matrix, meaning the rows (and columns) are orthogonal unit vectors. The inverse of an orthonormal matrix is simply its transpose, the matrix formed by changing the rows to

Chapter 11. Computational Geometry

11.4 Change of Basis in Vector-Matrix Form

187

columns. Using this information, the equation becomes, I2 M = J2 K2 O2

1 T

I I = I O2

I2 J2 K2 I =

I I I O2

x J2 x K2 x I2 y J2 y K2 y I2 J2 K2 z z z 0 0 0

0 0 0 1

where -O2 means [ -x, -y, -z, 1 ]. This is a compact, simple and efficient computation for M. The translation matrix is constructed simply by assigning the negative of O2 to the last row of a matrix that has been initialized to identity, and the transposed rotation matrix is formed by assigning the coordinates of I2, J2 and K2 as the first three columns of a matrix that has been initialized to identity. The transformation M is the product of these matrices. As an example, this procedure can be used to find the viewing transformation, V. The first coordinate system is XYZw, the world coordinate system, and the second is XYZe, the eye coordinate system. The resulting transformation is V, the viewing transformation. World coordinates will be transformed into eye coordinates. Given the view vector endpoints, EYE and AIM, one can find O2, I2,J2 and K2 using vector operations (see Figure 11-24). Yw Ye AIM Zw Xe Ze EYE Xw

Figure 11-24. Change of Basis for the Viewing Transformation. The process involves several steps: 1. K2 = || EYE - AIM ||, the unit direction vector of the Ze axis with respect to XYZw.

Chapter 11. Computational Geometry

11.4 Change of Basis in Vector-Matrix Form

188

The next step is to find I2, the direction of the Xe axis. It is necessary to specify a direction, often called the UP direction, which, when crossed with K2, forms the direction of I2. To follow the viewing conventions established during the trigonometric version of the viewing transfmoration discussed earlier, UP = J ([0,1,0,0]), the unit vector along the Yw axis. One must be careful, however, when UP is parallel to K2, i.e. when UP K2 = 0, for example, looking along Yw. In this case, one must specify another UP vector. To follow the conventions established in these notes, in this case specify I, [1,0,0,0], the unit Xw axis. IF ( J K2 ) approx.= 0 THEN I2 = I ELSE I2 = || J K2 ||

3. 4. 5. 6.

J2 = K2 I2

(note that J2 will be a unit vector because K2 is perpendicular to I2)

Create the transposed rotation matrix, R, from I2, J2, K2 as described earlier. Create the translation matrix, T = T(-AIM), because O2 = AIM in this case. Finally, compute V = T R.

As an example, consider the situation shown in Figure 11-25, with AIM = [0,0,0,1] and EYE = [1,0,0,1].
Yw Ye Xe Ze Xw

AIM
Zw

EYE

Figure 11-25. Example of Change of Basis Applied to the Viewing Transform. Following the steps, 1. K2 = || EYE - AIM || = || [1,0,0,1] - [0,0,0,1] || = [1,0,0,0].

Chapter 11. Computational Geometry

11.5 Instance Transformations

189

J K2 = [0,1,0,0] [1,0,0,0] = [0,0,-1,0] (i.e., -K). This is not zero length, so I2 = || J K2 || = [0,0,-1,0].

3. 4.

J2 = K2 I2 = [1,0,0,0] [0,0,-1,0] = [0,1,0,0]. Form the transposed rotation matrix, R: I2 R = J2 K2 I

0 = 0 1 0

0 1 0 0

1 0 0 0

0 0 0 1

Form the negative translation, T, noting that O2 = AIM: I 1 0 I T = = 0 1 I 0 0 O2 0 0 0 0 1 0 0 0 0 1

Finally, compute V = T R: 1 V = 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0 1 0 1 0 0 1 0 0 0 0 0 0 0 = 0 1 0 1 0 1 0 0 1 0 0 0 0 0 0 1

Look at the columns of the R and V matrices in the previous example. They are the unit direction vectors of XYZe with respect to XYZw. This means that given V, one can find the directions of the axes of XYZe.

11.5 Instance Transformations

The logic used in the derivation of the change of basis transformation also is useful in forming instance transformations for drawing objects that are represented in a local coordinate system. Given an object represented in its own local coordinate system, we want to draw it at a given position and orientation in world space by transforming the local coordinates to world coordinates as lines are drawn. This is especially applicable when using the CTM matrix of a 3D pipeline.

Chapter 11. Computational Geometry

11.5 Instance Transformations

190

For example, it is simple to compute the points on a circle with its center at the origin and with a unit radius. Refer to this as the canonical circle in its local coordinate system. Drawing an instance of a circle at a point on an arbitrary plane in 3D space is another problem, however. We would like to compute the local coordinates of the points on a circle and use an instance transformation to map them to points on the 3D circle in world coordinates. The problem is to find this transformation from the information describing the world position and orientation. The problem, as usual, is finding the transformation M that maps XYZL coordinates into XYZw coordinates. The derivation of this transformation is similar to the derivation of the change of basis transformation. First, compute the origin and direction vectors of the local coordinate system in world coordinates from the position and orientation information given for the object. Next, form M from these vectors like the change of basis derivation, using the equation, Pw = PL M: XL YL 1 = 0 0 ZL 0 O 0 1 0 0 0 0 1 0 0 0 M = M 0 1

This M is made by simple row assignments from the four vectors, the three unit direction vectors XL, YL, and ZL, and the origin vector O, all expressed in world coordinates. For example, compute the instance transformation to map a unit cylinder to the world position and orientation as shown in Figure 11-26. In the figure, the XYZL axes have been drawn Yw YL XL [0,2,2,1] ZL Basic cylinder in local coordinates Zw Instance of cylinder in world coordinates
YL XL ZL

axis of cylinder is [1,0,0,0] Xw

Figure 11-26. Cylinder Instance Transformation Example.

Chapter 11. Computational Geometry

11.6 Transforming Continuous Geometries

191

in world coordinates. By observation we find that O = [0,2,2,1], XL = [0,0,-1,0], YL = [0,1,0,0], and ZL = [1,0,0,0]. Next, these vectors are assembled into the instance transformation M: XL M = YL 0 = 0 1 ZL 0 O 0 1 0 2 1 0 0 2 0 0 0 1

Notice that the rotation sub-matrix is a simple RY(90). If the direction vectors could not have been found by observation, then the viewing transformation vector operations described in the previous section could have been used to compute XL, YL and ZL. Also, if the cylinder instance is to be scaled, such as changing the radius (or radii, e.g. making the cylinder elliptical) or length, then these transformations should be concatenated with the positioning and orienting transformation M to form the instance transformation. The scaling must be done in XYZL, before M.

11.6 Transforming Continuous Geometries

Transformations have been discussed in terms of points and lines. As one would expect, it is possible, and often desirable, to transform equations representing continuous geometry. For example, consider the plane: A x +B y +C z +D = 0. First re-write the algebraic form in matrix form: A B = 0 xyz1 C D Transform all points in the plane by the transformation M, mapping [x,y,z,1] [x', y', z', 1]: x' y' z' 1 = x y z 1 M The coefficients of the transformed plane, B , are found by post-multiplying the above equation by M-1B: The original plane equation is

Chapter 11. Computational Geometry

11.6 Transforming Continuous Geometries

192

P B = 0. The transformed points are P = P M, so, inverting to find P, P = P M-1. Substituting this equation for P into the original plane equation above, ( P M-1) B = 0. Re-group the matrix and B multiplication, P ( M-1 B) = 0. Or, B = M-1 B, and P B = 0 is the new plane equation. For example, consider the perspective projection of the plane y + z - 1 = 0, the plane along the projector through y = 1 (Figure 11-27). The plane equation is [x y z 1] [ 0 1 1 -1 ]T. The y y=1 y+z-1=0

x d=1 z Figure 11-27. The Plane y + z - 1 = 0. transformation M is the perspective projection matrix, P. For d = 1, this becomes: 1 0 A' B' = 0 1 C' 0 0 D' 0 0 0 0 1 0 0 0 1 1
1

0 1 0 1 = 0 1 1 0 0 1 0 0

0 0 1 0

0 0 0 0 1 = 1 1 1 0 1 1 1

So, the transformed plane equation is y - 1 = 0, or y = 1. This shows that the top clipping plane of the viewing volume transforms to a plane perpendicular to the y axis (Figure 11-28). Using similar logic, the coefficients of a transformed conic, i.e., circle, sphere, parabola,

Chapter 11. Computational Geometry

11.6 Transforming Continuous Geometries

193

y y+z-1=0 y-1=0 x d=1 z Figure 11-28. Original and Transformed Planes. ellipse, or hyperbola, can be derived. For simplicity, we will consider only two-dimensions in the following. The general 2D conic equation is, a20x2 + 2a11xy + a02y2 + 2a10x + 2a01y + a00 = 0. This can be put into matrix form, a 20 a 11 x y 1 a 11 a 02 a 10 x x a 01 y = x y 1 A y = 0 1 1 a

a 10 a 01 00 or, P A PT = 0

Transforming the curve is transforming all its points, so again apply the transformation M to [x, y, 1] to produce [x',y',1]. Inverting the equation to solve for [x,y,1] in terms of M-1 and [x',y',1], and substituting into the equation above, the transformed conic is found: A' = M-1 A (M-1)T. As an example, consider translating a unit circle at the origin by [1,0,0,0]. The A matrix equation becomes: 1 0 0 x x y 1 0 1 0 y = 0 0 0 1 1 1 0 0 A = 0 1 0 0 0 1

Chapter 11. Computational Geometry

11.7 Rotation About An Arbitrary Axis Through The Origin

194

Therefore, the transformed A matrix, A', is found as follows: 1 0 0 A' = 0 1 0 1 0 1

1 1 T 1 0 0 1 0 0 1 0 0 1 0 0 1 0 1 0 1 0 0 1 0 = 0 1 0 0 1 0 0 1 0 0 0 1 1 0 1 1 0 1 0 0 1 0 0 1

1 0 1 A' = 0 1 0 1 0 0 This is the equation: (x-1)2 + y2 -1 = 0, which, as expected, is the equation of a unit circle with center [1 0]. In three dimensions, the conics are ellipsoids, spheres, paraboloids, etc. The matrix form is x y = 0 xyzw A z w where the matrix A for 3D conics is 4 by 4. Exactly the same as the 2D case above, transforming the 3D conic by a transformation M creates a transformed conic represented by A' = M-1 A (M-1)T.

11.7 Rotation About An Arbitrary Axis Through The Origin

Using basic transformations one can derive the transformation matrix that rotates an object about an arbitrary unit axis, u, through the origin. y u ux z Figure 11-29. Rotation Axis Notation. uy x uz ux + u y + u z = 1
2 2 2

Chapter 11. Computational Geometry

11.7 Rotation About An Arbitrary Axis Through The Origin

195

The approach is to first perform a change of basis that places the object in a coordinate system in which u becomes the z axis. Then rotate the object by the angle about this z axis and transform the object back into the original coordinate system. The first transformation is precisely the viewing transform, hence derive it using the change of basis in vector-matrix logic. The direction K2 is u. Next compute I2: I2 = || ( J K2 ) || Now compute J2 = K2 I2 uz u x --------------------- 0 --------------------I2 = 2 2 2 2 ux + uz ux + u z ux u z J 2 = --------------------2 2 ux + uz
2 2 uy uz u x + u z --------------------2 2 ux + uz

Now create R, the transposed rotation transformation, using these as the columns: uz u x u y --------------------- --------------------- u x 2 2 2 2 ux + uz ux + uz R = 0 ux --------------------2 2 ux + uz ux + u z u y uy uz --------------------- u z 2 2 ux + uz
2 2

Form the composite transformation as (recall R-1 RT): M = R R(,z) RT uz ux u y --------------------- --------------------- ux 2 2 2 2 u x + u z ux + u z M = 0 ux --------------------2 2 ux + uz
2 ux 2 + uz

cos sin 0 u x uy uy sin cos 0 --------------------0 0 1 2 2 uy uz ux + uz --------------------- u z 2 2 ux ux + u z

uz --------------------2 2 ux + uz

ux --------------------2 2 ux + uz

uy u z 2 2 u x + u z --------------------2 2 ux + uz uy uz

Chapter 11. Computational Geometry

11.8 Review Questions

196

After considerable algebra, one arrives at a convenient form: u x vers + cos M = u u vers u sin x y z
2

u x u y vers + u z sin u x u z vers uy sin u y vers + cos

u y u z vers + u x sin uz vers + cos

ux u z vers + u y sin uy u z vers u x sin where vers = 1 - cos.

11.8 Review Questions

1. Given AIM = [3,2,1] and V below, compute the unit direction vectors of the eye coordinate system and EYE for d = 2. 0.9487 0 0.3162 3.1623 0.0953 0.9535 0.2860 1.9069 0.3015 0.3015 0.9045 0.6030 0 0 0 1

2. 3.

How can one determine if two parallel vectors are in the same or opposite directions? What would be the approximate values for s and t for these line-line intersection cases? P1 (t) P1 (t) P2 (s) P2 (s)

4. 5.

How would the condition line in plane be computed? Given a viewing transformation V, show how to compute the translation and rotation matrices for the change of basis formulation. Also, show how the AIM location can be found from V.

Chapter 11. Computational Geometry

Chapter 12. Hidden Line and Surface Removal

12.2 3D Screen Space

199

considerably more complex for concave objects than convex.

12.2 3D Screen Space

When discussing the 3D wireframe pipeline, perspective was derived as a projection transformation from 3D to 2D. In order to perform HLR and HSR, it must be possible to determine if a given point in 3D space is in front of or behind another along a given line of sight. It was shown how line of sight is satisfied by projection. If two 3D points project to the same [x,y] coordinates on the picture plane, then they lie along the same projector. Ye Projectors in XYZe (wsx,wsy,0) Xe Ze Zv Figure 12-3. Lines of Sight Before and After Projection. But, how can depth comparisons be performed after projection if the data is 2D? The answer is that, for HLR and HSR to determine visibility, one must retain the z coordinates of points after projection. The perspective projection transformation derived earlier, P, does create projected z coordinates for this purpose. However, some perspective projection transformations, namely those with the COP at the origin, do not yield z coordinates that can be used for depth comparisons. We refer to 3D screen space coordinates as XYZV. Xv Yv Projectors in XYZv

12.3 Hidden Line Algorithms

The most common HLR approach is to compare each edge with each face to see if it is all or partially obscured. Approaches vary in that some compare edges to faces and others compare faces to edges. The problem is essentially one of clipping a line against a polygon.

Chapter 12. Hidden Line and Surface Removal

12.3 Hidden Line Algorithms

200

One of the first such algorithms was Robert's HLR algorithm. The general HLR idea is: Step 1. Perform the viewing and projection transformations so that all lines are in the 3D screen space. Step 2. Compare each line with each polygon in 2D, i.e. using only xv and yv. There are three possible cases: 2a. The line is outside the polygon. Action: keep the line. Yv

Xv Figure 12-4. Line Outside a Polygon. Note that determining if a line is outside a polygon requires (1) determining if any line of the polygon intersects the given line, and (2) determining if the endpoints of the given line lie inside the polygon (known as the point containment test ). Examine the lines and polygon in Figure 12-5. Line AB has endpoints outside the polygon, but intersects it. Line CD has endpoints inside the polygon and also intersects it. Line EF is a simple case, the location of the endpoints can be used to classify the line as crossing. A C E D F B

Figure 12-5. Lines and a Polygon. 2b. The line is inside the polygon in 2D. Determine if the line penetrates the plane of

Chapter 12. Hidden Line and Surface Removal

202

sorted based on decreasing distance (depth) from the eye. Each polygon is filled, that is, the pixels within the projected boundaries of the polygon on the screen are written into the frame buffer, beginning with the polygon farthest from the observer and proceeding to the closest. Pixels within closer polygons are written after those in farther away polygons, and therefore overwrite them in pixel memory. In this case, the 3D screen space can be (h,v,zv), where h and v are the pixel coordinates computed using window to viewport mapping, and zv is the projected z coordinate. 1. 2. 3. Project polygons into 3D screen space, Sort polygons by zv values, Fill each in order, from farthest to nearest to the eye. 1 2 3

Figure 12-8. Overlapping Polygons in the Painters Algorithm. If two polygons overlap in 2D and intersect in zv, then depth sorting produces incorrect results. In this case, the two polygons must be split into non-intersecting polygons, which is a complex polygon clipping-like process. This greatly increases the complexity and diminishes the elegance of the algorithm. The speed of the Painters algorithm depends upon the filling speed of the graphics system because all polygons, visible or hidden, are filled. Filling is itself a computationally intensive process, so this is not the most efficient algorithm. It is, however, the simplest to implement.

12.4.2 Z-Buffer Algorithm

A more sophisticated approach, based somewhat on the Painters algorithm, is to maintain a buffer (array) that holds the depth (usually zv) for each pixel in the frame buffer. This depth

Chapter 12. Hidden Line and Surface Removal

12.4 Raster Algorithms

203

buffer, called a z-buffer, is consulted prior to storing a pixel in pixel memory. This also requires that the polygon filling algorithm compute a depth (z value) for each pixel. Pixel Memory pixel values pixel depths Z-Buffer

pixel h,v Figure 12-9. Z-Buffer Storage.

depth at h,v

When a polygon is filled, for each pixel inside it a depth is computed and compared to the depth stored in the z-buffer for the corresponding pixel in pixel memory. If the polygons pixel is nearer to the eye (greater-than for a right-handed system, less-than for a left-handed system), the new pixel depth is stored in the z-buffer and the pixel value is stored in pixel memory. If the new pixel is farther away, it is ignored. After all polygons have been processed, the pixels remaining are the nearest pixels, i.e. the visible pixels. Often, to avoid the excessive overhead of pixel-by-pixel graphics system calls, a separate array of pixel values is created for this algorithm. In this case, the picture is not drawn until all polygons have been processed. Then, all the pixel values are written at once to the display. The 3D screen space is the same as the previous, (h,v,zv). 1) 2) 3) Project polygons into 3D screen space. Initialize pixel and depth buffers (two dimensional arrays). FOR each polygon Pi DO FOR each pixel & depth inside Pi DO IF depth is closer than stored depth[i,j] THEN depth[i,j] = depth pixel[i,j] = pixel (or write the pixel to the screen directly) ENDIF ENDDO ENDDO 4) Draw picture from pixel[][] data (if necessary)

Chapter 12. Hidden Line and Surface Removal

12.4 Raster Algorithms

204

The z-buffer approach has been implemented in hardware and is used in many high performance graphics cards.

12.4.3 Watkins Scan-Line Algorithm

The Painters algorithm is simple to implement, but cannot deal with complex intersecting polygons and blindly fills every polygon whether invisible or not. The z-buffer algorithm handles more complex geometry but requires large amounts of memory (the pixel and depth buffers). Both algorithms compute the picture in random pixel order, meaning pixels are computed anywhere on the screen, one or more times. The Watkins algorithm was one of the first efficient scan-line hidden surface algorithms. The Watkins algorithm is designed expressly for display on a raster system and bases all computations on scan-lines and pixels. The entire scene is processed one scan-line at a time. For each scan-line, from the top of the viewport to the bottom, a horizontal plane, conceptually a scan-plane, is passed through the scene and intersected with all polygons. The intersection of a plane with a plane produces a line (or several lines for concave polygons), so the horizontal scan-plane produces a set of lines, called segments. These segments are horizontal lines with respect to Xv - Yv, but are in fact 3D lines in the 3D screen space. Yv Polygon B
a scan-line a b c d c d b

Polygon A Xv
Zv

Xv
scan-plane

Figure 12-10. Scan-line and Scan-Plane. Viewed in the scan-plane, the segments may overlap. Depth comparisons are made in the overlapping area(s) of segments to find which parts of segments are in front of others, i.e. visible. Overlapping segments are sub-divided and comparisons continue until no overlaps exist.

Chapter 12. Hidden Line and Surface Removal

12.4 Raster Algorithms

205

The resulting non-overlapping segments are rasterized into pixels and finally output to the display device. Then, we proceed to the next scan-line. A segment is rasterized by computing the locations of the individual pixels along the horizontal scan-line at integral h positions and writing these into pixel memory. Due to the properties of lines and polygons in the 3D screen space, the Watkins algorithm actually is edge-driven. That is, most computations are performed in 2D with some additional bookkeeping for the depth values. Watkins algorithm uses a divide and conquer (non-deterministic) algorithm for subdividing segments to remove overlaps. Simple segment conditions are processed, and other complex cases are sub-divided until they become simple. Scan-line coherence, the similarity between visible segment locations on one scan-line to the next, is used to eliminate unneeded subdivision. It is the scan-line coherence methodology and segment visibility resolving logic that distinguishes the Watkins algorithm from others, and makes it one of the fastest (and complex) HSR algorithms. The Watkins segment rasterization phase can be changed to produce a raster image with only the edges, not the interior of surfaces. This is done by outputting marker pixels at the ends of resolved segments and background pixels at the interior pixels. The Watkins algorithm can also output lines. Instead of rasterizing the resolved segments on a scan-line, the HLR version of Watkins remembers the segment boundaries, marked by the polygons that generate them, from one scan-line to the next. By maintaining a list of growing lines, adjacent scan-lines can be compared to detect the appearance, disappearance, and continuation of lines. Consider the following example:
scan-line 2 3 C B 4 5 6 7 Action New lines A and B. New line C, extend A & B. Extend A, B & C. Extend A, B & C. End A & C, Extend B. End B.

When comparing the lines to the segments on the current scan-line, there are 3 possibilities: 1. A segment boundary has no counterpart growing line.

Chapter 12. Hidden Line and Surface Removal

12.5 Polygon Clipping

206

Start a new growing line with start and end positions at the given point. 2. A segment endpoint on the completed scan-line matches a growing line. Extend the growing line by setting its endpoint to the given point. 3. A growing line has no counterpart on the completed scan-line. The line ends: output it to the display or file being created and remove it from the data structure.

12.5 Polygon Clipping

HLR and HSR computations require that polygons be closed. For example, the Painters algorithm projects the polygons into XYZv and then fills them. During this process, what should be done if an edge of a polygon lies outside the viewing frustum? Unfortunately, simply performing line clipping on the edges is wrong. Line-clipping algorithms are inadequate because the resulting polygon may not be closed, as illustrated in Figure 12-11. Polygon clipping is needed clipping boundary clipped polygon is no longer closed

Figure 12-11. Line Clipping Applied to a Polygon. to maintain closed polygons for later computations such as filling and point containment tests. Polygon clipping appears to be considerably more difficult than line clipping (Figure 12-12).

Chapter 12. Hidden Line and Surface Removal

12.5 Polygon Clipping

207

invisible regions are shaded

clipping boundaries Figure 12-12. Complex Polygon Clipping. The Sutherland-Hodgman algorithm [Sutherland] is an elegant approach to polygon clipping. The algorithm sequentially processes each edge around the perimeter of a polygon, testing the start and end vertices of the edge against the clipping boundaries. Given the start point of an edge, P1, and the endpoint, P2, there are four cases to consider as shown in Figure 12-13. The Invisible region is shaded

P2:output

I:output P2

P1 Case 1: P2 output. Case 2: I output.

P1 P2:second output I:first output Case 4: I & P2 output. P1

P2 Case 3: no output.

Figure 12-13. Four Clipping Cases for Polygon Clipping. algorithm processes each pair of edge endpoints, P1 and P2, clipping the edge against each boundary. Note that the process for each boundary is the same, i.e. as points are output from one

Chapter 12. Hidden Line and Surface Removal

12.5 Polygon Clipping

208

boundary, they can be processed by the next. Each boundary evaluation is identical to the others, so the algorithm can be programmed using re-entrant code, i.e., one routine that processes P1 and P2 points can call itself. A simpler approach is to clip the whole polygon against one boundary at a time, storing the output vertices in a second buffer (array of points). If any points are in the buffer after clipping on one boundary, then clip on the next boundary, this time reading points from the second buffer and outputting them to the first. Continue clipping on the remaining boundaries, reading from one buffer and storing into the other, toggling input and output each time. The buffers must be dimensioned at least twice the maximum number of vertices allowed per polygon. This can be understood intuitively by considering a polygon shaped like a saw blade. Figure 12-14 shows a pseudo-code illustration of the basic procedure for clipping a polygon against one boundary, specified by an argument of data type boundary that represents the plane coefficient vector [A,B,C,D].
ClipPolygon(boundary B): p1 = last vertex in input buffer; FOR( p2 = each vertex of input buffer (1st to last) ) DO IF State(p1, B) != State(p2, B) THEN I = intersection of line p1-p2 with the boundary B; Append I to the output buffer; ENDIF IF State(p2, B) = IN THEN Append p2 to the output buffer; p1 = p2; ENDFOR

Figure 12-14. Pseudo-Code Polygon Clipping Routine. The routine State in Figure 12-14 returns the state of the given point with respect to the given plane, either IN or OUT. An example of State in pseudo-code is shown in Figure 12-15.
State( Vertex P, boundary B): B < 0 THEN IF P State = OUT ELSE State = IN; ENDIF END

Figure 12-15. An Example of the State Function.

Chapter 12. Hidden Line and Surface Removal

12.6 Scan Converting Polygons

209

For homogeneous polygon clipping, there would be seven planes, starting with WZERO. For example, consider Figure 12-16. The initial vertex buffer would contain the four boundary B P2 invisible side of boundary P3

I23 I30 P1

P0 Figure 12-16. Example Polygon and Boundary. ordered points { P0, P1, P2, P3 }. After processing by the code in Figure 12-14, the second vertex buffer would contain the five ordered points { I30, P0, P1, P2, I23 } that represent the clipped polygon. Other approaches to polygon clipping have been published [Liang][Vatti].

12.6 Scan Converting Polygons

Scan converting a 2D polygon whose vertices have been mapped to screen (pixel) coordinates involves computing the locations of all pixels whose centers lie inside or on the exact polygon boundary, commonly called filling. This can be done by scanning the polygon one infinite horizontal scan-line at a time (y=Yscan), from top to bottom (or bottom to top). The 2D scan converting algorithm consists of the following steps. For each scan-line, Yscan: 1. Compute the intersections of the scan-line with the edges of the polygon. A convex polygon will produce one pair of intersection points, while a concave polygon may produce a large number of pairs.

Chapter 12. Hidden Line and Surface Removal

12.6 Scan Converting Polygons

210

Grid locations are the coordinates of pixels. In some systems, this is the center of a pixel. In others, this may be a corner.

Xv Figure 12-17. Pixel Grid and Polygon. 2. Sort these intersections by increasing x. Due to the arbitrary ordering of vertices around a polygon, the intersections will not generally be ordered left to right (increasing value). 3. Group the ordered intersections into pairs to form scan-line segments that lie inside or on the polygon. This is commonly called the even-odd rule. For any closed object, as an infinite line crosses its boundaries, it toggles from outside to inside.
9 8 e1 7 e2 6 5 e3 4 e4 3 2 1 2 segments

Yv
1 segment

scan-line A

scan-line B scan-line C

Xv
1 2 3 4 5 6 7 8 9

Figure 12-18. Segments Formed by Intersections of Scan-Lines and a Polygon. 4. Rasterize each of the segments by marching horizontally along the segment from (Xstart, Yscan) to (Xend, Yscan) assigning pixel values at pixel locations, i.e. grid coordinates.

Chapter 12. Hidden Line and Surface Removal

12.6 Scan Converting Polygons

211

Exact segment x coordinates must be rounded to the nearest integer pixel coordinates. Care must be taken to round properly. In general, the start (left) segment x should be rounded to the next larger integer value, and the end (right) segment x should be rounded to the next smaller integer value, as shown in Figure 12-19. The library routine CEIL(X) provides the Pixel center locations 1 2 3

-1

4 Xv

Xstart = -0.8 exact line endpoint coordinates

Xend = 3.8

Figure 12-19. Segment Coordinate Rounding for Rasterization.

rounding up function and FLOOR(X) provides the rounding down function.

12.6.1 Vertex Intersections

It is possible for a scan-line to intersect a vertex if the vertex y coordinate is on a scan-line Yscan. This will always happen for integer coordinates, and can happen even if real coordinates are used. One way to handle vertex intersections is to examine the previous and next vertices to determine if the vertex is a local minimum or maximum. If so, count the vertex intersection as two intersections with the same x coordinate. Otherwise, count it as one. This is further complicated by horizontal edges on scan-lines. prev vertex

prev next

Figure 12-20. Vertex Intersections Along a Scan-Line. Another approach is simply not to allow vertices to have y values on a scan-line by

Chapter 12. Hidden Line and Surface Removal

12.6 Scan Converting Polygons

212

adjusting any offending y values, or all y values, prior to scan-converting the polygon. This requires that coordinates be stored as real (or rational integer) values. If a y value is on a scan-line, then add a small amount to its value to place it between scan-lines. The small amount should be large enough to avoid scan-line intersections but small enough to insure that the rounded values fall on the same scan-line. Some approaches simply add 0.5 to all rounded y coordinates. Because the scan-converting process finds pixels to the nearest integral x and y coordinate anyway, image quality will not suffer (for our purposes, anyway).

12.6.2 3D Scan Conversion

To provide pixel depths for z-buffer algorithms, the 2D scan-converting algorithm must be extended to include pixel depth computations. Now consider the scan-line as an edge-on view of an infinite plane: Y= Yscan, as shown in Figure 12-21. This changes the scan-line - edge intersections from 2D to 3D, intersecting the plane Y= Yscan with the 3D segment between two 3D vertices of the polygon. The result of each intersection calculation is now (Xi, Yscan, Zi), where Xi Yv

scan-plane Xv Zv Figure 12-21. Segments and the Scan-Plane. is the X coordinate as before, but Zi is the Z value at the intersection. This list is sorted by Xi as before, and paired into segments. The segments are now 3D lines in the plane Y=Yscan, from (Xstart,Yscan,Zstart) to (Xend,Yscan,Zend). As the line is rasterized into integral X pixel addresses, the corresponding z depth at the rounded pixel x values must be computed (Figure 12-22). This z depth can be used in

Chapter 12. Hidden Line and Surface Removal

12.7 Scan-line Z Buffer Algorithm

213

z-buffer comparisons. intersections @ Yscan 1 2 x1 x2 z1 z2 Zv (x1,z1) pixel center locations Xv (x2,z2)

Figure 12-22. Z Interpolation Along a Segment.

12.7 Scan-line Z Buffer Algorithm

Storing an entire screen of pixel values and pixel depths can be very costly and may not be necessary. Consider instead computing the image one scan-line at a time, and outputting each fully resolved scan-line to the output device in display order. The z-buffer depth comparisons will still be used to do hidden surface removal, but only along one scan-line. This is the scan-line z-buffer algorithm: Map and clip polygons to XYZv FOR each scan-line Yscan DO Clear the scan-line; FOR each polygon Pi DO Compute sorted intersection segments Sj of Pi on Yscan; FOR each segment Sj DO Rasterize Sj into pixel values and pixel depths; Compare and replace pixels in z-buffer with pixels from Sj; ENDDO ENDDO Draw the scan-line; Clear the scan-line means to set all pixels to BACKGROUND and all depths to FAR AWAY.

Chapter 12. Hidden Line and Surface Removal

12.8 References

214

12.8 References
1. 2. 3. Sutherland, I. E. and Hodgman, G. W., Reentrant Polygon Clipping, Communications of the ACM, Vol. 17, No. 1, 1974. Liang, Y. and Barsky, B. A., An Analysis and Algorithm for Polygon Clipping, Communications of the ACM, Vol. 26, No. 11, November, 1983. Vatti, B. R., A Generic Solution to Polygon Clipping, Communications of the ACM, Vol. 35, No. 7, July, 1992.

12.9 Review Questions

1. 2. What is the "3D screen space" ? Be specific. Explain the process of back-plane rejection. For what type of data is back-plane rejection valid? For what type of data will backplane rejection completely remove hidden lines? 3. Give a description, in step-by-step outline form, of the two hidden surface (polygons) algorithms named below. Provide sufficient detail to clearly show the flow of the algorithm, and to describe what needs to be done at each step: 3A. the "Watkins Scan-line" algorithm, 3B. the "Scan-line Z-Buffer" algorithm. 4. Given an ordered list of polygon vertices with an unspecified normal direction, describe how to compute if this polygon is convex or concave (non-convex). 5. Describe the contents and storage needs for the z-buffer for (1) the Z-Buffer HSR algorithm, (2) the Scan-line Z-Buffer algorithm. 6. Given an eye coordinate system with the COP at the origin and image plane at z = -d, what are the projected z coordinates? Is this projection useful for HLR and HSR?

Chapter 12. Hidden Line and Surface Removal

215

Chapter 13. Realistic Display Effects

During the discussion of hidden surface removal, we often referred to the polygons color and other operations for storing the pixel value for the visible pixel belonging to an object. Suppose we have prepared 3D data consisting of a set of polygons covering a sphere, transformed the data to the screen space, and have performed HSR. We have assigned the color gray to each polygon in the data and the pixel values were stored in pixel memory holding the final image. Unfortunately, such an image would appear as a flat gray circle (with straight sides), without perceivable 3D depth characteristics, as illustrated in Figure 13-1. The problem is that we have

Figure 13-1. HSR Rendered Image. ignored color variations due to lights, geometry and material. We have determined the visible pixels in the image, but have not assigned color values to the pixels that indicate 3D effects. To create a realistic image of 3D objects requires more than just hidden surface determination. It also requires simulation of the interaction of light with the geometry and surface properties of the objects that determines the light that reaches our eyes. Recalling the discussion of color, light is visible radiation that interacts with surfaces in complex ways. Incident light from many sources irradiates the surface of an object of a particular material. Light is absorbed by, reflected from, and possibly transmitted through, the surface. This behavior is determined by the geometry of the surface and its material properties, including its color. The terms illumination model, lighting model and reflectance model are used to describe models for the interaction of light and geometric objects. For our purposes, we will use the term illumination model for the mathematical model used to compute the color of a pixel corresponding to a visible point on the surface of an object. The shading model combines an illumination model

Chapter 13. Realistic Display Effects

13.1 Illumination Models

216

with methods for varying the color across the surface of objects in an efficient manner.

13.1 Illumination Models

Illumination models are based on the physical behavior of light and materials and involves complex energy transfer phenomena that have been the subject of study for hundreds of years. Early illumination models attempted to capture the gross behavioral characteristics of light and materials to produce computer generated images that would appear realistic to the eye. These models were somewhat perceptually based, in that they were developed empirically by experimentation to produce visually pleasing results using simple mathematical formulae to model mathematically indescribable physical behavior. We begin by considering the fundamental components of a basic illumination model, diffuse, specular, and ambient reflection. For the time being, consider the incident light to be achromatic light that can be characterized by an intensity. Later we will consider more complex models for light.

13.1.1 Diffuse Reflection

Micro-facets, microscopic plates on the surface of an object, reflect the incident light in all directions. For objects with dull or matte surfaces, the micro-facets on the surface create a distribution of reflected light that is concentrated in the direction of the surface normal, as shown in Figure 13-2. This is characteristic of diffuse reflectance. The intensity of the reflected diffuse Incident light direction L Surface Point in question N reflected light as f()

Figure 13-2. Diffuse Light Reflection.

Chapter 13. Realistic Display Effects

13.1 Illumination Models

217

light can be approximated by the cosine curve shown in Figure 13-2, where L is the unit direction vector from the point in question toward the light, N is the unit surface normal vector, and is the angle between L and N. We know that cos = L N if L and N are unit vectors. The intensity of reflected light for diffuse reflection is Id = IL Kd cos(), where IL is the incident light, Kd is a surface property representing how much incident light is reflected diffusely from a given surface. Usually, Kd is a value ranging from 0.0 to 1.0.

13.1.2 Specular Reflection

For polished surfaces, bright spots or highlights, appear when the light source, surface normal and the eye are in the right positions to capture some of the specular reflection. Looking at an ideal reflector, i.e. a mirror, a ray of incident light would be reflected only along the direction RL, which would be seen only if the eye were located exactly as shown in Figure 13-3. If the N incident light RL Eye

Figure 13-3. Specular Light Reflection. surface is not a perfect reflector, as most are not, then the reflected light is scattered around RL and the intensity diminishes as the eye moves away from the direction of perfect reflection, RL. There are two common approaches to simulating specular reflection. The first, and least commonly used today, involves the angle between the ideal reflection direction, RL, and the vector to the eye, E, as shown in Figure 13-4. The specular component of reflected light is Is = IL Ks cosn. where Ks is a surface property representing the amount of incident light specularly reflected from

Chapter 13. Realistic Display Effects

13.1 Illumination Models

218

N RL E Eye

Figure 13-4. Specular Reflection Vectors. a surface, n is the specular exponent, a surface property representing the distribution of specularly reflected light about the reflection vector RL (values usually are 2 to 100), and is the angle between RL and E. In this case, it is computed using cos = RL E if RL and E are unit vectors. Given L and N, the vector RL is computed by sweeping L through N by an angle . N B L A B RL A=(N L)N B = A - L = (N L) N - L RL = A + B = 2 A - L RL = 2 ( N L ) N - L

Figure 13-5. Reflection Vector RL Computations. Note that it is possible to have RL E < 0, which cannot be allowed. This means cos() < 0, so > 90 , which makes it impossible for reflection. In this case, the specular reflection should be set to 0. Instead of using the RL vector for computing the specular component, a more popular approach is to define another vector, H, which is the unit bisector of L and E as shown in Figure 13-6. H is computed using vector operations: H = || E + L || assuming E and L are unit vectors.

Chapter 13. Realistic Display Effects

13.1 Illumination Models

219

This simplifies the computations considerably and eliminates the logical problems of RLE<0. N H L E Eye

Figure 13-6. H Vector Computations. If H and N are unit vectors, cos = H N and the specular component of reflected light is Is = IL Ks cosn. This is the most common method for specular reflection.

13.1.3 Ambient Light

We may be tempted at this time to formulate the illumination model by simply summing the diffuse and specular components of reflected light, i.e. I = Id + Is or I = IL Kdcos() + IL Ks cosn Using only diffuse and specular reflection makes the objects appear unrealistically sharp and stark, as though the they were in a black room without walls. In a real environment, light is transmitted from lights and reflected from objects onto other objects in a very complex manner described by laws of physics. To account for these inter-object light reflections in a simple way, most illumination models include an ambient component, called ambient light. Ambient light has no incoming direction and is constant across each surface. The purpose of ambient light is to compensate for complex lighting conditions that cannot be simulated (or are too time-consuming to simulate). Ambient light produces constant intensity surfaces, Ka Ia, where Ka is a surface property

Chapter 13. Realistic Display Effects

13.1 Illumination Models

220

representing how much surrounding ambient light a particular surface reflects, Ia is a global illumination property, representing the amount of ambient light in the lighting environment surrounding the objects. In this case, it is the intensity of ambient light in the environment.

13.1.4 Ambient - Diffuse - Specular Model

Combining ambient, diffuse and specular reflection yields: I = Ka Ia + IL Kd cos + IL Ks cos n or, I = Ka Ia + IL Kd ( L N ) + IL Ks ( H N )n This is a basic lighting model used in many computer graphics applications. Typically, the values for the surface properties Ka, Kd and Ks are scaled so that: Ka + Kd + Ks <= 1.0. Logically, this limits the total amount of reflected light to be no greater than the incident light. Also, if a surfaces coefficients add to less than one, it will reflect less light than a surface with a larger sum. To model the reflection of colored light, we again are faced with two modeling approaches: the physically-based simulation of light interacting with surfaces, or the early empirical solution that utilizes the tri-stimulus theory of color. The simulation approaches model colored light as a specular curve and represent the surface material properties as specular functions instead of the constant coefficients that we have seen. The reflection models operate on the specular curves to produce the net reflected light as a specular curve. See [Hall] for a discussion of this process. A simpler approach is based on the tri-stimulus theory of color. Light is modeled as intensities of primary component colors, a trichromatic vector, typically RGB. The incident light, I, in the previous equations changes from an intensity to a vector, R I = G B

Chapter 13. Realistic Display Effects

13.1 Illumination Models

221

Likewise, the surface model coefficients become coefficient vectors, K a, red K a = K a, green K a, blue K d, red Kd = K d, green K d, blue K s, red K s = K s, green K s, blue

The full color lighting model is now a vector equation: I = Ka a IL Kd (L N) IL Ks ( H N )n The products of trichromatic vectors, for example Ka a above, are computed by multiplying corresponding array elements to produce a vector as the result. This is sometimes called "array multiplication." Similarly, the product IL Kd (L N) above is a vector product multiplied by a scalar, producing a vector as a result.

13.1.5 Light Sources

The basic illumination model utilizes two sources of light: ambient light and incident light. Ambient light has no direction and is a user-defined "fudge-factor" to account for inter-surface reflections. Incident light, however, has a direction that depends upon the type of light source. There are a number of different models for different types of light sources. Each light source model contains a color of the source light, position and direction information, and parameters which control the behavior of the light. The color is either an achromatic intensity (typically assumed to be white light) or a trichromatic vector. The simplest and most often used is the point light source, like a light bulb. A point light source is modeled as a location and color. The incident light direction is computed using the world coordinates of the point in question and the light position: L = || PL - P || where P is the point in question and PL is the position of the point light source. Another useful light source is parallel light, such as sunlight. Parallel light is represented by a direction vector. The light direction vector, L, is simply the negation of this vector, and is the same for all points on all objects. The spotlight source is representative of the many other types of light source models. Typically, a spotlight source is represented by a position, direction, cone angle and attenuation

Chapter 13. Realistic Display Effects

13.2 Polygon Shading Methods

222

function. The direction of incident light is computed similar to the point light source, using the spotlight position and the point in question. The intensity of the incident light, however, is varied according to the geometry of the spotlight direction, spotlight cone angle and the incident light direction using the attenuation function. These computations are similar in nature to the computations for diffuse reflection, where the angle between the incident light direction and spotlight direction determines the factor applied to the spotlight color to determine the incident light intensity. The light is a maximum along the spotlight direction, falling off to a minimum value at the spotlight cone angle.

13.1.6 Multiple Light Sources

To model a lighting environment with more than one light source, one executes the illumination model for each light individually and sums the results. It is usually necessary to "clamp" intensities to a maximum value, such as unity, during this summation. During lighting computations, when a light intensity exceeds unity, it is set ("clamped") to unity.

13.2 Polygon Shading Methods

13.2.1 Flat Shading
For scan-line and z-buffer hidden surface processing of polyhedral objects, i.e., objects described by planar polygons, a single shade (color and intensity) can be computed for each polygon using an appropriate lighting model. Typically, the light source is considered to be sufficiently far from the polygons surface so that the rays are essentially parallel. Therefore, any point on the polygon can be used to compute the L and E vectors and all pixels for this polygon

Chapter 13. Realistic Display Effects

13.2 Polygon Shading Methods

223

have the same value. This is called flat shading.

Figure 13-7. Flat Shaded Object.

13.2.2 Smooth Shading

Rendering shaded images of objects with curved surfaces is often done by approximating the surfaces by some number of planar polygons. This is called tessellation and the polygons are called facets. This presents special problems to rendering methods that use flat shading. The human eye easily discerns small changes in intensities, so one easily sees that is an image of a sphere made of many flat polygons. Henri Gouraud developed a shading model for rendering tessellated objects by interpolating the intensities across the boundaries of the polygons. Gouraud shading, or intensity interpolation, involves computing intensities at the vertices of polygons. An example of smooth shading is shown in Figure 13-8. If the normals at the vertices can be computed from surface

Flat shaded cylinder

Smooth shaded cylinder

Figure 13-8. Examples of Smooth Flat and Smooth-Shaded Cylinders. equations, such as the equations of a cylinder as shown in Figure 13-8, then these normals are stored at the vertices and used for the lighting computations. When the surface equations are not

Chapter 13. Realistic Display Effects

13.2 Polygon Shading Methods

224

known, as is the case in systems that approximate all surfaces with planar polygons, then the shared vertices where facets from the same surface meet, the geometric normals are averaged to form a lighting normal that will be used for intensity computations. The averaged normal at a shared N2 8 3 N1 1 5 2 4 6 NA 7

N12

Figure 13-9. Normal Averaging for Smooth Shading. vertex is the vector average of the plane normals of all polygons sharing the vertex. In Figure 139, for example,
8

N12 = || || N1 || + || N2 || || and NA =

i=4

Scan-line computations must now linearly interpolate the color of pixels based on the intensity or trichromatic color components across a scan-line segment and along the edges from one scan-line to the next. For example, in Figure 13-10, the colors of pixels along the indicated scan-line is a linear interpolation of the colors IL and IR, which are themselves linear interpolations along the edges I1I2 and I1I3, respectively: IL = I1 + t12 ( I2 - I1 ) and IR = I1 + t13 ( I3 - I1 ), where t12 and t13 are parameters that vary from 0 to 1 along each edge. The intensity across segment IL-IR is interpolated with a parameter s that varies from pixel to pixel across the scan-line: Is = IL + s ( IR - IL ). Unfortunately, Gouraud shading produces noticeable mach bands of intensity patterns. These are due to discontinuities in the intensity variations across polygon boundaries. B. T. Phong suggested averaging the normal itself, instead of the intensity. This reduces the

Chapter 13. Realistic Display Effects

13.2 Polygon Shading Methods

225

I1 t12 IL
XL,ZL,IL

t13 IR
XR,ZR,IR

scan-line

I2 I3 I4 Figure 13-10. Intensity Interpolation While Filling a Polygon. bands produced by Gouraud shading. Notice the nonlinear relation between the intensity I and the normal N in the lighting model equation. Phong shading, or normal interpolation, produces smoother images, but requires more complicated per-pixel computations. The full lighting model must be computed for each pixel.

Chapter 13. Realistic Display Effects

13.3 Ray Tracing

226

13.3 Ray Tracing

Ray tracing (also called ray casting) is a versatile and elegant method for rendering images with realistic display effects. Our eyes integrate the light that falls on the retina from all directions to form the image we see. This is difficult to compute, so instead consider a single ray of light passing out of the eye through each pixel, asking the question What contributed to light along this ray that reached the eye through this pixel? The idea is to trace or cast lines from the eye position through each pixel on the image plane into space. Whatever is seen along the ray is displayed as the pixel color on the raster Scene Data Yw Ye Ray intersections Zw Ze Eye Figure 13-11. Ray Tracing. screen. Ray tracing requires the computation of the intersections of a ray with the geometry describing the scene, typically planes, but also spheres, cylinders, and other geometric forms. For each ray, all intersections with all objects are computed. Line of sight visibility is determined along a ray easily: the ray intersection nearest to the eye is the visible point, called the "visible hit." When the ray is modeled as a parametric vector equation of a line, such as P + t V for t > 0, the ray parameter t is the distance measure along the ray that determines the nearest hit. The ray tracing process is basically simple: Xw Xe Screen Ray

Chapter 13. Realistic Display Effects

13.3 Ray Tracing

227

Procedure RayTrace: FOR each pixel on the screen in world coordinates DO Compute the vector ray equation through the pixel; Set the "visible_hit" parameter to "none"; FOR each object in the scene DO Compute the ray intersection(s) with the object; IF an intersection is nearer than visible_hit THEN set visible_hit to new intersection parameters; ENDIF ENDFOR IF there was a visible_hit THEN Set the pixel color using the illumination model; ELSE Set the pixel color to the background color; ENDIF ENDFOR The ray tracing process simplifies and enhances the rendering process considerably: 1. There is no need to transform the geometry into screen coordinates because rays can be cast just as easily in world coordinates. 2. Projection is a by-product of the process because each ray is a projector. If the rays emanate from the eye point, then the resulting image will have perspective projection. If the rays are all parallel and perpendicular to the screen, the image will appear to be orthographically projected. 3. Clipping is not needed because rays are cast only through pixels on the screen. In addition, intersections with geometry at or behind the eye, i.e. with zero or negative ray parameters, are easily ignored as invisible. 4. The world coordinates of the intersection point and the surface normal at this point are available for illumination model computations and more sophisticated rendering effects. 5. More complex geometry than simple polygons can be used because the computations only require the intersection with a ray. Before we abandon scan-line and z-buffer rendering, however, there is a significant disadvantage to ray tracing: it is many times more computationally expensive. Ray tracing is

Chapter 13. Realistic Display Effects

13.3 Ray Tracing

228

elegant, but very time-consuming, compared to scan-line methods. Indeed, much of the research in ray tracing is directed towards improving the computation times.

13.3.1 Shadows
One rendering effect that adds considerable realism to an image is shadows. Shadows occur when light from a light source is obscured from portions of the surfaces of objects by other objects. We have simply ignored this phenomenon so far. Shadows modify the behavior of the illumination model. Ray tracing facilitates shadow computations. After the visible hit point, called Q from now on, has been computed from the ray emanating from the eye, another ray is cast from the Q toward each light source. If this shadow ray intersects any object before reaching the light (see Figure 1312), then this light source cannot contribute any light reflection toward the eye for this pixel. In a B A shadow ray

Ye Q Xe Ze Screen

Figure 13-12. Shadow Ray Vectors. ray tracing program, the shadow ray is "just another ray" to be intersected with the geometry of the scene. It is possible to have the same function process the initial ray and the shadow ray.This contributes to the elegance of the process. The shadow ray, however, must be processed differently than the initial ray from the eye. The ray parameter for the initial ray is only used for relative distance comparisons, so its magnitude is generally unimportant. It is a true ray with a semi-infinite parameter range (t > 0). The magnitude of the "V" direction vector in the initial ray equation is not important. For the shadow

Chapter 13. Realistic Display Effects

13.3 Ray Tracing

229

ray, however, it is necessary to know if any object intersects the ray between Q and the light. This means that the shadow ray is really a segment, with upper and lower parameter bounds. Typically the shadow ray equation is: P(t) = Q + t (PL - Q) for 0 < t <= 1. If the visible hit point is determined to be shadowed from a light source, then only ambient light reflects from the point. Therefore, the illumination model is truncated to the ambient portion for this light for this point: I = Ka Ia.

13.3.2 Reflection and Refraction

The illumination model has considered only light reflected directly from lights to objects to the eye. In a real environment, however, light reflects from surfaces to objects as well. Such global illumination computations add to realism at increased computational cost. After computing the initial ray intersection, Q, one can ask the question: "What light could have reached the eye as a result of reflection from a surface at Q?" The question is answered by casting a ray in a direction R corresponding to ideal reflection with the direction E to the eye (Figure 13-13). This ray simulates reflection and becomes a separate incident light path to trace.

Ye E Ze

N Xe Screen

Figure 13-13. Reflection Ray Vectors. The direction R is found with vector computations similar to the specular reflection vector RL discussed in section 13.1.2: R = 2 ( E N ) N - E. If the object is transparent, that is, light can be transmitted through it, a ray can be cast

Chapter 13. Realistic Display Effects

13.3 Ray Tracing

230

through the surface to simulate the refracted light reaching the pixel. The direction of the transmitted ray, or refraction ray, is determined by Snells law of refraction that relates t, the angle between the transmitted ray T and the normal N, and nt, the coefficient of refraction of the transmitted medium, to i, the angle of the incident ray and ni, the coefficient of refraction of the incident medium (Figure 13-14): ni sin i = nt sin t. N E i ni (air) nt (other)

Figure 13-14. Vectors for Refracted Light. Each reflected and transmitted ray is further processed as another incident ray, in turn spawning its own reflections and refractions. The result is a ray producing process that creates a binary tree of rays, all to produce the color for one pixel. The resulting data structure resembles a tree of intersection data nodes (Figure 13-15). Once the tree has been computed, the resulting Node Data: surface properties (Ka,Kd,Ks,n), lighting model geometry (Q,N,E,L), shadow results per light.
T

Visible Hit R

Ii
R T R T

It Figure 13-15. Ray Tree for a Single Ray.

intensity (color) must be computed for this pixel. This is done, typically using a recursive

Chapter 13. Realistic Display Effects

13.4 Ray Geometry

231

procedure, for each data node (bottom up), using a formula of the form: Inode = Ki Ii + Kr Ir + Kt It where: Ii is the intensity computed with the nodes data for the incident ray and the light sources using an illumination model, Ir It Ki, Kr, Kt is the intensity from the reflection ray branch of the tree, is the intensity from the transmission ray branch of the tree, are the surface properties that determine the contribution of each ray (and sum less than or equal to 1.0). In theory, parallel and perfectly reflecting objects can create an infinite number of rays. In practice, the surface properties are used during the ray casting process to attenuate (decrease) the "intensity" of each ray at each intersection. When the ray intensity of one of the rays (reflection or refraction) decreases below a certain limit, no new rays are cast. Alternatively, some ray tracing programs simply stop casting rays after a certain tree depth is reached.

13.4 Ray Geometry

Ray tracing is performed in world space. The problem becomes finding the pixel locations in world coordinates (Figure 13-16). Let S be the position in XYZw of the lower left corner of the Ye
AIM

window-screen Xe Ze EYE

Yw DY S Xw Zw Figure 13-16. Vectors for Ray Geometry. window. The viewport (screen) pixels must be mapped onto a grid in this window. S then becomes DX

Chapter 13. Realistic Display Effects

13.4 Ray Geometry

232

the starting pixel location. (The other corners could do as well, of course.) Let the unit direction vectors of the eye coordinate system in XYZw be called xe, ye, ze. Given these, AIM, and wsx and wsy (the window half-sizes), S is found using vector operations in the image plane in XYZw space. S = AIM - wsx xe - wsy ye Let DX be the world vector between the centers of two adjacent pixels on the same scanline. That is, DX is a horizontal step. Given wsx, wsy and the viewport resolutions (number of pixels across and down the screen), DX can be found as shown in Figure 13-17. With S, DX and

L 2wsx DX = ------------------------- = ------------------------- x e ( Res x 1 ) ( Res x 1 )

2wsy DY = ------------------------- y e ( Res y 1 )

Figure 13-17. DX and DY Vector Computations. DY, it is possible to march across the screen pixel by pixel, tracing rays at each location. The ray equation is P(t) = P + t V. For perspective projection, all rays start at the eye, so P = EYE. Each ray passes through a pixel, so these are the two points on the line. The direction vector V = (Pi - EYE) where Pi is the given pixel location. For orthographic projection, P is formed with the same [x,y] coordinates as Pi and a z coordinate that is appropriate for clipping points behind the eye. In other words, select the z coordinate of P such that ray intersections with the ray parameter t <= 0 are invisible. All rays are parallel and perpendicular to the screen, so V = -K = [0,0,-1,0], a constant. Advancing to the next pixel on a scan-line is a computation using the vector DX: Pi+1 = Pi + DX for i = 0..Resx-1

Chapter 13. Realistic Display Effects

13.5 Ray-Polygon Intersections

233

Advancing to the next scan-line can be done using one of two methods. Given that Pi is at the last pixel on a scan-line, advance to the next by returning horizontally to the first pixel location, then moving up to the next scan-line: Pi+1 = ( Pi - 2 wsx xe ) + DY Another approach is to save the location of the first pixel on the current scan-line, thereby avoiding the computations above. Call this Si. Advance to the scan-line now simply by computing: Pi+1 = Si + DY Si = Pi+1

13.5 Ray-Polygon Intersections

Given a ray, P(t) = P + t V, and a planar polygonal surface through a point P0 and a normal N satisfying ( P - P0 ) N = 0, we first compute Q, the point of intersection of the ray and the infinite plane. Next we must determine if Q lies inside the bounded surface, a polygon in 3D space. It is assumed that the boundary consists of an ordered set of straight line segments between vertices. This procedure is commonly called the point containment test, or the inside/outside test. Typically, the plane vertices are stored in a list data structure of some kind, ordered consistently with the given normal. For example, the vertices traverse the boundary in a CCW orientation when viewed down the normal, or something similar. 4 1 2 3

Figure 13-18. Concave and Convex Polygons and Vertices Forming Their Boundaries. First consider point containment for convex polygons.

Chapter 13. Realistic Display Effects

13.5 Ray-Polygon Intersections

234

13.5.1 Point Containment for Convex Polygons

If the polygon in question is convex, then its geometric properties can be used to simplify and speed the point containment test. The interior region of convex polygons can be described as the intersection of half-spaces formed by the edges (sides). This can be expressed as if a point is inside each half-space formed by the infinite extension of each edge then the point is inside the polygon. For efficiency, invert the rule to if a point is outside any half-space (etc.), then the point is outside the polygon.

P2 V2

P0 V0

V1 P1

Figure 13-19. Convex Polygon as the Intersection of Half-Spaces Formed by Lines. This is the basis for the quick-rejection algorithm shown in Figure 13-20. FOR each edge PiVi DO if [(Q-Pi)Vi]N > 0 THEN RETURN OUTSIDE;

END RETURN INSIDE; Figure 13-20. Point Containment Algorithm for Convex Polygons.

13.5.2 Semi-Infinite Ray Point Containment Test

One technique for determining point containment for non-convex polygons is the semiinfinite ray test. Cast a ray in the plane from Q in any convenient direction and count the intersections with the boundary segments. This process is time-consuming and is complicated by

Chapter 13. Realistic Display Effects

13.5 Ray-Polygon Intersections

235

intersections with vertices. The algorithm in Figure 13-22 simply restarts if the ray intersects a vertex.

3 Ray

Figure 13-21. Semi-Infinite Ray for Point Containment Tests.

RESTART: Set random ray direction in plane; count = 0; FOR each edge of polygon DO IF ray intersects edge THEN IF intersection is at a vertex THEN GO TO RESTART; ENDIF count = count + 1; ENDIF ENDDO IF count IS ODD THEN RETURN "INSIDE" RETURN "OUTSIDE"

Figure 13-22. Semi-Infinite Ray Point Containment Algorithm.

13.5.3 Sum of Angles Point Containment Test

Another somewhat more efficient technique for non-convex polygons is the sum of angles method. This is based on the property that if a point is inside a closed polygon, the sum of the angles between vectors from the point to successive pairs of ordered vertices will be 360 degrees (2). Using Vi to mean the vector from Q to Pi, note that: (Vi Vi+1) N Vi Vi+1 = = |Vi | |Vi+1| sin |Vi| |Vi+1| cos

( Vi Vi + 1 ) N sin so, tan = ----------- = ----------------------------------------cos Vi Vi + 1

Chapter 13. Realistic Display Effects

13.6 Orienting Lighting Normals

236

Pn-2 Q Yw Pn-1 Vi Xw P3 Vi+1

P1
i

Figure 13-23. Sum of Angles Algorithm Notation.

total = 0 FOR each vertex Pi of the polygon DO /* including Pn-1 to P0! */ Vi = Pi - Q Vi+1 = Pi sine =
+1

- Q

(Vi x Vi+1)

cosine = Vi Vi+1 IF |sine|0 AND |cosine|0 THEN RETURN INSIDE /* on vertex */ = ATAN2(sine,cosine) IF THEN RETURN INSIDE /* on edge */ total = total + ENDFOR IF |total | 2 THEN /* a large tolerance, +/- 0.5, can be used here */ RETURN INSIDE ELSE RETURN OUTSIDE ENDIF

Figure 13-24. Sum of Angles Point Containment Algorithm.

13.6 Orienting Lighting Normals

Depending on the form of the data representing objects, it may be necessary to determine which side of a surface is "visible to the observer." Similar to the discussion for back plane rejection, the data may be closed or open. Closed data means that it is not possible to see the back of a surface, so surface normals are "oriented" and can be used directly as lighting normals in the

Chapter 13. Realistic Display Effects

13.6 Orienting Lighting Normals

237

illumination model. Open data may be oriented or not, meaning that surface normals may be used to determine the front or back of a surface, or not. For example, for a single plane, as the eye position is rotated around the data, the plane normal will sometimes face the observer and other times face away from the observer, determined by the result of E N. For un-oriented data, the proper surface normal must be selected for the lighting normal when the viewing position changes. In effect, this makes each surface a thin solid. Consider an edge-on view of the visible surface at the point of ray intersection, Q, as shown in Figure 13-25.

Eye a Ea

Na La Q

Light a

Surface Lb Light b

Eb Eye b Nb

Figure 13-25. Possible Normals for Different Eye Positions for a Surface. Consider the different combinations of the two different eye positions (labelled "Eye a" and "Eye b"), two light positions ("Light a" and "Light b") and two different surface normals ("Na" and "Nb"). The objective is to select the normal, Na or Nb, to give the proper lighting effect. There are 4 cases to consider depending on the eye and light locations. Select a lighting normal Nl from the surface normal, N, as follows:
Nl = current surface normal IF( E N < 0 ) THEN "flip lighting normal: Nl = - Nl" ENDIF IF( L Nl < 0 ) THEN "eye hidden from light: use ambient light only" ELSE "light is visible to eye: use the full lighting model" ENDIF

Figure 13-26. Lighting Normal Algorithm.

Chapter 13. Realistic Display Effects

13.7 Intersecting a Ray with a Convex Polyhedron

238

13.7 Intersecting a Ray with a Convex Polyhedron

The interior of a convex polyhedron (a volume whose bounding faces are convex polygons) can be described as the intersection of half-spaces bounded by the infinite planes formed by each of the faces. Similar to the point containment test for convex polygons, the algorithm for determining if a point is on a convex polyhedron involves testing on which side of each face the point in question lies. Given a ray, P(t) = P + t V, one approach to finding the points of intersection is: 1. Compute the set { ti } of intersections with the infinite planes of the object. For a parallelepiped, i = 0..5; for a tetrahedron, i = 0..3. 2. Classify each point { Qi } = { P + ti V } by forming the homogeneous dot product with the other planes Bj, i.e., Qi Bj, where Bj is a homogeneous plane vector [ A, B, C, D ]T, and Bj has been formed such that [A,B,C] is the right-hand outward normal of the face. The result of the dot product will classify the point: Qi Bj Qi Bj Qi Bj 3. = > < 0 0 0 point on plane, IN ( or ON ), point is on +normal side, definitely OUT, point is on -normal side, maybe IN.

After classification, there will be: 1. 2. no points in { Qi } that classify as IN, in which case the ray misses the polyhedron, two points in { Qi } that classify as IN. In this case, let Q1 be the point with the smallest ti, the enter point, and Q2 be the other, the exit point. The normal for the enter point is the plane normal; the normal for the exit point is the negative of the normal. A second approach is even simpler and faster. Observe that if the ray intersects the convex

polyhedron, the entering t value must be less than the exiting t value. Conversely, if the ray intersects the infinite planes made by the faces of the object outside of the face boundaries, the entering t value will not be less than the exiting value. The entering intersection will have VN < 0, and the exiting VN > 0 (see Figure 13-27). The algorithm shown in Figure 13-28 combines the two steps in the previous algorithm into

Chapter 13. Realistic Display Effects

13.7 Intersecting a Ray with a Convex Polyhedron

239

t1, Q1

N2 (exit) t2, Q2

N1 (enter) P(t) = P + t V Figure 13-27. Ray Intersecting a Polyhedron. one loop.

Initialize tenter = -, texit = +. FOR each plane Bi of the object: IF V Bi 0 THEN IF P Bi > 0 THEN no hits, exit ELSE skip to the next plane ENDIF ENDIF t = - P Bi / V Bi IF V Bi < 0 AND t > tenter THEN tenter = t, Benter = Bi. IF V Bi > 0 AND t < texit THEN texit = t, Bexit = Bi. ENDFOR IF tenter <=texit AND tenter != - THEN RETURN 2 hits: tenter on Benter, normal +Benter and texit on Bexit, normal -Bexit ELSE RETURN "no hits". ENDIF

Figure 13-28. Ray - Convex Polyhedron Intersection Algorithm.

Chapter 13. Realistic Display Effects

13.8 Intersecting a Ray with a Conic Solid

240

13.8 Intersecting a Ray with a Conic Solid

The general equation of a conic solid is a x2 + b y2 + c z2 + d x + e y + f z + g xy + h yz + i xz + j = 0. This equation can be represented in matrix form: x y = 0 xyz1 A z 1 where A is a matrix of constant coefficients (the values a-j). For example, a sphere of radius R and center at the origin has the equation x2 + y2 + z2 - R2 = 0 and has the following A matrix: 1 0 0 0 0 1 0 0 A = 0 0 1 0 0 0 0 R2 Letting P = [x,y,z,1], the equation is, P A PT = 0. The intersection of a ray with a general conic can now be computed using this more convenient matrix form. The ray equation is the familiar P(t) = P + t V. Note that all coordinates are assumed to be homogeneous, so V = [ x, y, z, 0 ]. Substituting P(t) for P in the matrix equation above yields: ( P + tV ) A ( P + tV )T = 0. After some manipulations, this reduces to: t2 ( VAVT ) + t ( PAVT + VAPT ) + PAPT = 0. This is a quadratic formula in an unknown parameter t. Letting a = VAVT, b = PAVT + VAPT, and c = PAPT, yields the standard form of the quadratic formula: a t 2 + b t + c = 0 with the well known solution:

Chapter 13. Realistic Display Effects

13.8 Intersecting a Ray with a Conic Solid

241

b b 4ac t = ------------------------------------2a The real roots of this equation, if any, are the parameter values of the intersection points of the ray and the conic. The discriminant inside the radical, b2 - 4ac, determines the solution: 1. if it is negative, there are no real roots and hence no intersection between the ray and conic, 2. if it is 0, there are two identical real roots, so the intersection is a point of tangency between the ray and conic, 3. if it is positive and non-zero, there are two real roots corresponding to the two intersection points between the ray and the conic. Generally, two identical roots can be handled the same as two distinct roots. The ray-conic intersection routine simply checks the sign of the discriminant and returns no hits if it is negative or 2 hits at t1 and t2 if it is positive. The values for t1 and t2 are the minimum and maximum of the two roots, respectively. To find the normal vector to a conic at a given point, Q, one begins with the basic equation of the transformed conic. Letting P = [x,y,z,1] and A be the transformed coefficient matrix, PAP
T

= f ( x, y, z ) = 0

The normal vector is found using the vector gradient operator, , defined as: N = f ( x, y, z ) = fi + fj + fz x y z

Applying the gradient operator to the conic at point Q: N = ( PAP )

T T T T = ( PAP )i + ( PAP )j + ( PAP )k x P=Q y z

P=Q

The only variables involved in the partial derivatives are x, y and z in P. For example, evaluating

Chapter 13. Realistic Display Effects

13.9 Ray-Tracing CSG Solids

242

the x partial derivative,

T T P P ( PAP ) = ( AP ) + ( PA ) x x x T

P = 10 0 0 = i x
T T T ( PAP ) = i ( AP ) + ( PA )i x

= 2 ( PA )i Similarly,

T T ( PAP ) = 2 ( PA )j y T T ( PAP ) = 2 ( PA )k z

Summing the gradient, N = ( PAP ) = ( 2 ( PA )i )i + ( 2 ( PA )j )j + ( 2 ( PA )k )k Dropping the factor of 2 (N is normalized anyway) and combining the vector components into a matrix, the result is: N = QA The w coordinate of N must be set to 0.
T T T T

13.9 Ray-Tracing CSG Solids

CSG (Constructive Solid Geometry) is a method for representing volumetric solids as primitive solids combined using the Boolean set operations Union, Difference and Intersection. Consider the following example of set operations applied to 2D areas: Letting TRUE mean inside or on and FALSE mean outside, construct truth tables for the operators as shown in Figure 13-30. In 3D, this logic applies to volumes. For example, in Figure 13-31 a hole is placed through a block by subtracting a cylinder from it. The truth tables, however, are the same. Each primitive must be able to answer the query: is the point [x,y,z] inside you ? CSG can be used with ray

Chapter 13. Realistic Display Effects

13.9 Ray-Tracing CSG Solids

243

Union B

Difference

Intersection

A-B

Figure 13-29. CSG Set Operations Illustrated with 2D Areas. UNION A T F B T T T F T F A T F T F F DIFFERENCE B F T F A INTERSECTION T F B T T F F F F

inside A OR inside B AB

inside A AND outside B A ~B

inside A AND inside B AB

Figure 13-30. Truth Tables and Geometric Meaning of Set Operations.

Figure 13-31. Boolean Difference Applied to Volumes. tracing to create realistic images of complex objects composed of Boolean combinations of simpler primitives. This involves: 1. computing the intersections of the ray with each of the primitives and storing them (parameter value, object identifier, etc.), 2. 3. sorting the intersections by increasing ray parameter value, evaluating the resulting intersection list using the CSG expression, i.e., the list of primitives and the Boolean operators applied to them: primitive1 + primitive2 - primitive3 + ... primitiven To evaluate the intersections using the CSG expression, consider the ray as the real line starting at the left at - (far behind the eye), and proceeding to the right to + (far in front of the eye) as illustrated in Figure 13-33.

Chapter 13. Realistic Display Effects

13.9 Ray-Tracing CSG Solids

244

CSG object: Ye

ray

cylinder intersections CSG expression: box-cylinder Ze Eye Xe Screen box intersections

Figure 13-32. CSG Expression and Ray Intersections.

Figure 13-33. Ray Intersections Shown as Points on the Real Line. For each primitive, its current ray state is maintained as a Boolean variable Si for i = 1..#objects. Initially, assume the ray begins outside all primitives, so Si = FALSE. If it is possible for the ray to start inside a primitive, then additional computations are needed to determine the initial Si for each primitive. Begin processing each intersection along the ray, nearest to farthest (left to right on the real line), evaluating the Boolean expression using the current ray states of each primitive, Si. As each intersection is reached, denoted by the parameter tj, toggle the ray state of the corresponding primitive, Sj. Toggling means to change TRUE (IN) to FALSE (OUT), and viceversa, corresponding to entering the primitive on the first hit and leaving it on the second. The process terminates when: 1. 2. the expression evaluates to TRUE (IN), this is the visible intersection, the last intersection is passed without evaluating to IN, there is no intersection with the CSG object. (Consider the ray that intersects only a negative object.) This is illustrated in Figure 13-34. Given the individual ray states of the primitives, evaluating the Boolean expression utilizes the truth tables. Consider the CSG expression as a list of primitives, each with a SENSE of "+" or "-". An algorithm for evaluating the expression is shown in Figure 13-35.

Chapter 13. Realistic Display Effects

13.9 Ray-Tracing CSG Solids

245

Assign initial S values to each primitive FOR each intersection Ii in ascending ray parameter order DO Si = NOT Si IF EvaluateCSGObject() == TRUE THEN IF ray parameter value of Ii < 0.0 THEN RETURN "no hit: behind the eye" ELSE RETURN "visible hit is intersection Ii" ENDFOR RETURN "no hit"

Figure 13-34. CSG Ray Tracing Logic.

BOOLEAN EvaluateCSGObject: result = OUTSIDE FOR each primitivei DO IF SENSE(primitivei) == "+" THEN result = result OR Si IF SENSE(primitivei) == "-" THEN result = result AND (NOT Si) ENDFOR RETURN result

Figure 13-35. Evaluation of a CSG List with Primitive States.

Chapter 13. Realistic Display Effects

13.10 Texture Mapping

246

13.10 Texture Mapping

Texture mapping is a method for adding realism to rendered raster images without adding geometric complexity. Examples are wood grain, marble effects, rough surfaces on oranges and golf balls, decals (lettering and figures) on object surfaces. There are two main approaches to texture mapping: 1. 2. texture mapping (also called pattern mapping) bump mapping

13.10.1 Texture (Pattern) Mapping

A texture is a 2D array of texture elements, called texels, that are indexed by texture coordinates, (s,t). The texels and texture coordinates form a 2D space that is called texture space (Figure 13-36). t texel at (s,t) t

Figure 13-36. Texture Space. Texture mapping involves relating texture coordinates to pixels during the rendering process. Given a pixel resulting from rendering an objects surface, the color of the pixel can be computed from the related texel value in several ways, for example: 1. The texel value becomes the pixel value. For example, the texture map is an array of pixels from a digitized picture and the picture is mapped onto the surface.

Chapter 13. Realistic Display Effects

13.10 Texture Mapping

247

The texel value is a scale factor applied to the pixel value. For example, the texture map is an array of intensity factors that alter the color pattern of the surface to create the desired effect.

The texel value is an opacity (transparency) value or factor that determines the opacity of the pixels of the surface.

Usually, the texture coordinates are assigned in some way to a surface and are interpolated during rendering. For example, the texture coordinates of points on a sphere can be related to the two angles of spherical coordinates, the texture coordinates of a parametric surface can be the parameters of the surface, and so on. For polygons, texture coordinates could be assigned to each polygon vertex based on some desired effect. Then, the texture coordinates, (s,t), for vertices would be clipped with the edges of the polygon during polygon clipping, and interpolated during rasterization. For scan-line rendering algorithms, such as z-buffer hidden surface removal, the texture coordinates would be linearly interpolated along the edges of polygons and across segments during scan conversion. For ray-tracing, the texture coordinates of the visible ray intersection point, Q, would be interpolated from the surface or object vertices. When ray-tracing triangles, for example, barycentric interpolation is used to linearly interpolate the texture coordinates of any point Q inside a triangle from the texture coordinates at the 3 vertices. One problem with texture mapped images is aliasing, the appearance of jagged edges of objects. This occurs because the mapping of texels to pixels is highly non-linear, and one pixel may map to one or more texels as illustrated in Figure 13-37.

t Yw Xw s texture space Zw world space pixel space

Figure 13-37. Mapping of Texels to Pixels.

Chapter 13. Realistic Display Effects

13.10 Texture Mapping

248

During rendering, the texture coordinates of a certain pixel are computed with some interpolated method as discussed above. These coordinates are real numbers resulting from the pipeline transformations, clipping, and pixel by pixel computations. From these transformations, it is possible also to compute the area of the pixel in texture space. As shown in Figure 13-38, the relative area of the shaded pixel being rasterized and its center location has been mapped to texture space. The pixel center lies inside the given texel, but the pixel area maps to several texels.

Texels

Pixels

Figure 13-38. Pixel to Texel Mapping.. The texel value for a given pixel, therefore, can be computed in several ways depending upon the desired effect and accuracy. For example, the simplest approach is to use the texel value of the texel coordinates. Another approach is to use a simple or weighted average of the texels covered by the area. This is more time-consuming, of course.

13.10.2 Bump Mapping

Bump mapping is a texturing approach that uses geometric operations on the lighting parameters instead of mapping texels. It is the preferred method for adding surface variations to objects such as golf balls, fluted columns, etc. After the visible pixel location has been identified on an object via HSR, and prior to executing the reflectance model, bump mapping perturbs one or more of the lighting parameters to achieve the desired effect. The lighting parameters are the position (Q), object normal (N), light direction (L), eye direction (E), and reflectance vector (H).

Chapter 13. Realistic Display Effects

13.11 Review Questions

249

To perturb the normal vector, N, for example, to produce bumps on a golf ball, the object parameters of the hit point Q are used to alter the normal direction. The object parameters could be the two spherical angles of the point Q. The normal would be recomputed using functions based on the spherical angles such that "dimples" appear on the surface of the object. This is also called procedural texture mapping.

13.11 Review Questions

1. Given Q (point on the surface) = [1, 1, 0], PL (light position) = [-1, 1, 0] and N (normal at Q) = [0, 1, 1], compute the parametric vector ray equation of the shadow ray that can be used in a ray tracing program to determine if the light can contribute to the intensity at Q. For what range of values of the shadow ray parameter will intersections with other objects cause Q to be in a shadow? 2. Given the lighting situation: EYE = [1, 0, 1], PL (light position) = [-1, 1, 0], Q (point on the surface) = [1, 1, 0] and N (normal at Q) = [0, 1, 1], compute the unit vectors L, H, and E. 3. Describe, in step-by-step outline form, how one does smooth shading of a curved surface approximated by polygons (i.e. a sphere) using the scan-line zbuffer HSR algorithm. Make illustrations, and show all equations pertaining to smooth shading. 4. Given AIM = (1,0,1), EYE = (0,0,1), window half-size = 1, and horizontal and vertical (full screen) resolutions = 101, find: (a) Xe, Ye and Ze, the unit axes of the eye coordinate system expressed in XYZw (If you can't find them, assume values and continue; (b) the position vector for the lower left corner pixel on the window; (c) DX, the vector between adjacent pixels on a scan-line; (d) DY, the vector between adjacent vertical pixels between scan-lines. First show symbolic equations, then compute numeric values; (e) For ORTHOGRAPHIC projection, find the vector equation of the ray through S.

Chapter 13. Realistic Display Effects

13.11 Review Questions

250

A tetrahedron is a closed object bounded by four triangles intersecting at the four corner points. If the points are labelled A, B, C and D, then the triangles are ABC, BDC, ACD and ADB. Given the four corner points of a tetrahedron, A=(1,0,1), B=(2,0,3), C=(2,2,2), D=(3,0,1), the lighting model coefficients Ka = 0.2, Kd = 0.3, Ks = 0.4, and n = 2, the eye located at infinity on the +z axis, and the light located at infinity on the +x axis: compute the intensity of light from each face of the tetrahedron that reflects light to the eye using the ambient-diffuse-specular lighting model with the given parameters. Show equations in symbolic form prior to performing arithmetic.

For the ray tracing situation shown below: (a) compute and draw L, the unit direction to the light source; (b) compute and draw E, the unit direction toward the eye; (c) compute and draw R, the unit direction of ideal reflection; (d) compute cos(), and label on the diagram; (e) compute cos() and label on the diagram. First show symbolic equations, then compute numeric values. Light [-2,3] N [-1,1] eye [-4,1] Y Q= (-2,1) X

Chapter 13. Realistic Display Effects

14.1 Parameterization

251

Chapter 14. Curves

14.1 Parameterization
When storing the definition of a line, it is not necessary to store any intermediate points, just the endpoints. P1 P2

Figure 14-1. A Line Segment. P( t ) = P1 + t ( P2 - P1 ) or P( t ) = (1 - t ) P1 + t P2

Similarly, to draw an arc or circle, you only need the center C and radius R:

R C Figure 14-2. Circle. x( t ) = R cos( t ), y( t ) = R sin( t ) for 1 <= t <= 2 To draw the circle, one would connect a series of points around the circle by straight lines. The number of lines could vary from circle to circle to provide the appropriate quality of display. A general 3D curve can be described parametrically, for example, a helix (Figure 14-3): x( t ) = R cos( t ) , y( t ) = R sin( t ) , z( t ) = C t

It is possible to derive other parametric forms for analytic curves, such as second degree equations, i.e. conics: circles, ellipses, parabolas, and hyperbolae. For example, a circle can be represented by the parametric equations in Figure 14-4. Unfortunately, the parameter range of t needed to cover the whole circle is < t < . This is impractical and also shows that the point distribution around the circle is highly non-linear. It is computationally impractical to use a parameter range including the value infinity. By introducing a one dimensional homogeneous representation for the parameter, a more computationally convenient range is obtained. Replace t in the preceding equations with the ratio

Chapter 14. Curves

14.1 Parameterization

252

Figure 14-3. Helix. y t=0 R 2tR x = ------------1 + t2 x ( 1 t 2 )R y = --------------------1 + t2 t = - t= Figure 14-4. Parametric Equations for a Circle. T --- . The resulting parameterization is more notationally complex as it now involves two U T parameters, T and U. Represent t = as the ratio of -0 The complete circle is now traced in two parts. The top half is completed by holding U=1 with -1<=T<=1. The bottom half is then traced by holding T constant at 1 and letting U vary over the range -1 to 1, as shown in Figure 14-5. Comparing the parametric representation for curves, C(t) = [(x(t) y(t) z(t)], to the implicit representation, C(x,y,z) = 0, there are advantages and disadvantages for each. Parametric representation is best for drawing curves and operations requiring an ordered point set and for transformations of curves. Implicit representation is best for geometric intersection computations and for determining if a given point is on a curve. Ideally, one would like to have both the parametric and implicit representations. However, this is a very complex process at times and may not be possible at all for some curves. Using homogeneous coordinates: 0 R 1 wx wy w = 1 t t 2 2R 0 0 0 R 1

t = -1

t=1

Chapter 14. Curves

14.2 Spline Curves

253

y -1<= T <= 1 U=1 R x T=1 -1 <= U <= 1

T t = --U 2UTR x = ------------------U2 + T2 ( U 2 T 2 )R y = --------------------------U2 + T2

Figure 14-5. Homogeneous Parameterization of a Circle. The mathematical process of converting a parametric form into its corresponding implicit form has been termed implicitization. Finding the parameter value corresponding to a given (x,y,z) coordinate on a parametric curve is termed inversion. These processes are important computational geometry methods used in computer-aided design (CAD) systems.

14.2 Spline Curves

The presentations so far have been relatively simple shapes and their defining equations. To be of any practical value in applications, the methods discussed here must be extensible to more free-form curve shapes. Drafts-people have been designing free-form curves for years in 2D, using a flexible beam called a spline and weights called ducks. The spline is formed using as many ducks

255

For cubic polynomials (n=3), a convenient matrix notation is often used: a0 b0 a1 b1 a2 b2 a3 b3 A power basis vector coefficient matrix c0 d0 c1 d1 c2 d2 c3 d3

x y z 1 = 1 t t2 t3 T

A curve is defined by its coefficient matrix A. The goal is to define A in an intuitive and concise manner in order to produce a curve of the desired shape. One way to accomplish this is to force the curve to interpolate (pass through) four points spaced along the curve at pre-assigned parameter values. Given four known points, including their respective parameter values along the intended curve, we can derive the A matrix for the curve. Given P0 = ( x0, y0 , z0 ) , . . . , P3 = ( x3 , y3 , z3 ) and ( t0 , t1, t2 , t3 ), we have a system of 12 linear equations in 12 unknowns, and therefore we can solve for the coefficient matrix A. We know that P(ti) = Pi for i = 0..3, so we can form the matrix equation: x 0 y0 x 1 y1 x 2 y2 x 3 y3 This can be solved for A:
2 3 1 t0 t0 t0 2 3 1 t1 t1 t1 A = 2 3 1 t2 t2 t2 2 3 1 t3 t3 t3 1

z0 1 z1 1

2 3 1 t0 t0 t0

2 3 1 t1 t1 t1 = A 2 3 z2 1 1 t2 t2 t2 z3 1 1 t t2 t3 3 3 3

x0 y0 x1 y1 x2 y2 x3 y3

z0 1 z1 1 z2 1 z3 1

Given A, we can step along the cubic curve segment using sequentially increasing values of t (t0 <= t <= t3) to compute a set of points that can be connected by straight lines (Figure 14-7).

Chapter 14. Curves

14.3 Polynomial Forms

256

P1 t1 P0 t0 Pi=[xi,yi,zi]

P2 t2

Figure 14-7. Cubic Polynomial Interpolation through Four Points. Note that moving any one of the four points or changing its associated parameter value will change the coefficients, thus altering the shape of the whole curve (Figure 14-8). We now begin to see the concept of curve sculpting. This will not always produce a nicely-behaved or smooth curve P1 P2

Figure 14-8. Effects of Moving a Point on a Curve. because wiggles can result when a curve is forced to interpolate unevenly distributed points. One difficulty encountered with this formulation is that of matching slopes at the junction point (sometimes called a knot) between segments when creating a spline curve. 0 s=0 1 curve 1: P1(s) = S A1 knot 3 s=1 t=0 5 4 curve 2: P2(t) = T A2 6 t=1

Figure 14-9. Two Curves Meeting at a Knot Point. For many applications, a more appropriate and intuitive approach is achieved by specifying two endpoints and two parametric tangents. Taking the derivative of the curve with respect to the parameter: d P ( t ) = P' ( t ) = x ( t ) y ( t ) z ( t ) = 0 1 2t 3t 2 A dt We now use the curve endpoints (P0 and P1) and the end parametric tangents (P'0 and P'1)

Chapter 14. Curves

14.3 Polynomial Forms

257

to derive the coefficient matrix. Assuming t varies in the range [0,1]: P(0) 1 0 = P(1) = 1 1 P0 P ( 0 ) 0 1 P ( 1 ) 0 1 P1 P1 where: P' 0 = P' ( 0 ) = P' 1 = P' ( 1 ) = resulting in, 1 A = 1 0 0 0 1 1 1 0 1 0 2 0 1 0 3
1

0 1 0 2

0 1 A 0 3

d P(t) dt d P(t) dt

t=0

t=1

x0 y0 z0 1 x1 y1 z1 1 x0 x1

1 = 0 y0 z0 1 3 2 y1 z1 1

0 0 3 2

0 1 2 1

x0 y0 0 0 x1 y1 1 x0 y0 1 x y 1 1

z0 1 z1 1 z0 1 z1 1

This is known as Hermitian interpolation. Slope continuity between adjacent curves can now be directly controlled using a common tangent vector at the end of one curve segment and at the start of the next. P0 = P1(0)
t=1 P1(s) s=1 s=0 P2(t) t=0

P2 P2 = P2(1)

P0 = P1(0)

P1 = P1(1) = P2(0)

Figure 14-10. Slope Continuity between Spline Curves. We can now sculpt curves by moving one of the two endpoints or by changing one of the two tangent vectors. Note that a tangent vector has both magnitude and direction, and that changing either value alters the curve definition (see Figure 14-11). We can now apply a transformation matrix, M, to all points on a curve: [ wx wy wz w ] = [ x y z 1] M = T A M = T A'

Chapter 14. Curves

14.3 Polynomial Forms

258

changing direction

changing magnitude

Figure 14-11. Effects of Changing Parametric Tangent Magnitude and Direction. Now, instead of processing individual straight line segments through the 3D wireframe pipeline, we are able to transform an entire curve. The transformed curve is represented by the transformed coefficient matrix, A'. Using homogeneous coordinates, this can even include perspective.

Chapter 14. Curves

14.4 B zier Formulation

259

14.4 Bzier Formulation

Another popular curve form is the Bzier curve, shown in Figure 14-12. A degree n Bzier { bi }

b1
x

bn-1 control polygon

b0 t=0 Figure 14-12. Bzier Control Points and Curve.

bn t=1

curve is defined by a set of (n+1) ordered control points. When connected by straight lines, the points form what is commonly called a control polygon. The control points act somewhat like the ducks of the old drafting spline, however, only the first and last control points lie on the curve. The remaining points simply act as weights, pulling the curve towards each point. The resulting curve is said to mimic the overall shape of the control polygon, which is one reason Bzier curves are popular for interactive curve design. The general nth degree Bzier curve is defined as follows.
n

P( t) = where Bn,i(t) are Bernstein basis functions,

Bn, i(t) b i
i=0

n! B n, i(t) = -------------------- t i ( 1 t ) n i i! ( n i )! n is the degree of the curve (largest exponent of t), and bi, i = 0.. n are the control points. n! The constant factor in the Bernstein basis function is the binomial coefficient: -------------------- . i! ( n i )! It is expanded for some values of n and i in Figure 14-13. Recall that 0! = 1. Notice that for i = 0 and for i = n, the binomial coefficient is 1.

Chapter 14. Curves

14.4 B zier Formulation

260

i 0 n 1 2 3 4 1 1 1 1

1 1 2 3 4

1 3 6 1 4 1

Figure 14-13. Binomial Coefficients. A cubic curve is formed when n = 3 (i.e., 4 control points).
3

P(t) =

B3, i(t) bi =
i=0

b 0 B 3, 0(t) + b 1 B 3, 1(t) + b 2 B 3, 2(t) + b 3 B 3, 3(t)

= b0 ( 1 t ) 3 + b 1 ( 3t ( 1 t ) 2 ) + b 2 (3 t 2 ( 1 t ) ) + b 3 t3 This can be represented in matrix form: 1 P ( t ) = 1 t t2 t 3 3 3 1 = TM be b where: T Mbe [b] is the (power) polynomial basis vector, is called the Bzier basis change matrix, is the matrix of Bzier control points. 0 3 6 3 0 0 3 3 b 0 0 0 b1 0 b2 1 b 3

If more control over the shape of a curve is desired, additional control points may be added (increasing the degree of the polynomial). Using more than one curve segment can also provide additional control, especially in cases where the curve must pass through several points. One can show that the tangents at the ends of a Bzier curve are parallel to the first and last legs of the control polygon, respectively. Thus, in general for slope continuity, or C1 continuity,

Chapter 14. Curves

14.4 B zier Formulation

261

the next-to-last control point on curve #i, the common endpoint to curve #i and #i+1, and the second control point on curve #i+1 must be collinear (Figure 14-14).

curve i+1 curve i

Figure 14-14. Slope Continuity via Colinear Control Points. Second parametric derivative (curvature or C2) continuity between curve segments can also be achieved by proper placement of the third and third-to-last control points. Suppose we want a parametric cubic polynomial representation for a quarter circle in the first quadrant (Figure 14-15). P 3 = [0,1] @ t = 1 P 2 = [cos60,sin60] = [.5,.866] @ t = 2/3 P1 = [cos30,sin30] = [.866,.5] @ t = 1/3

P0 =[1,0] @ t = 0 Figure 14-15. Interpolation Points for a Quarter Circle. For point interpolation, the equation [P] = T A becomes: 1 0 0 1 1 0 0 0 0.866 0.5 0 1 = 1 0.3333 0.1111 0.0370 A 0.5 0.866 0 1 1 0.6667 0.4444 0.2963 0 1 0 1 1 1 1 1 Solving for A: 1 0 01 A = 0.04423 1.6029 0 0 1.4856 0.1615 0 0 0.4413 0.4413 0 0 Now find the Bzier control points for the same curve. To do this, first examine the matrix

Chapter 14. Curves

14.5 B-Spline Formulation

262

equation for the two curve forms, equating the power basis polynomial form with the Bzier form: P (t) = T A = T Mbe [ b ] We see that A = Mbe [ b ], so [ b ] = [ Mbe ]-1 A. For this example, 1 0 1 1 -3 1 = M be A = b 2 1 -3 1 1 0 0 1 0 0 1 1 0 0 1 0.04423 1.6029 0 0 = 1.015 0.535 0 1 1 1.4856 0.1615 0 0 0.535 1.015 0 1 -- 0 3 0 1 0 1 0.4413 0.4413 0 0 1 1 0 0

Examine the Bzier control polygon on the quarter circle: Y b3 =[0,1]

b2 = [0.535,1.015] b1 = [1.015,0.535] X b0 = [1,0]

Figure 14-16. Bzier Control Points for the Quarter Circle Example.

14.5 B-Spline Formulation

Another popular curve formulation is the B-spline representation. Like the Bzier formulation, the B-spline formulation also involves control points. In addition, however, the Bspline formulation involves parametric knots which give more degrees of freedom for controlling shape. The general mth order B-spline curve is defined as the summation:
m1

P( t) =

i=0

N m, i ( t )P i

where the B-spline basis function, Nm,i (x), is defined recursively:

Chapter 14. Curves

14.5 B-Spline Formulation

263

( x xi ) ( xi + m x ) N m, i ( x ) = ------------------------------------N m 1, i ( x ) + ------------------------------------ N m 1, i + 1 ( x ) ( xi + m 1 xi ) ( x i + m xi + 1 )

N 1, i ( x ) =

1 if ( x [x i,x i + 1] ) 0 if ( x [x i,x i + 1] )

where: xi are the parametric knots, m is the order of the curve (degree + 1), t is the parameter. The array of knots, called the knot vector, is a non-decreasing sequence of parameter values that gives local control to the shape of the curve. This means that moving a control point only affects the curve near the point, not the whole curve (as does the power basis polynomial formulation). From the B-spline basis definition, one sees that for a given parameter value, only the adjacent m knots affect the shape. Unlike the Bzier formulation, it is possible to vary the order of the curve, m, without varying the number of control points. As the order decreases, the curve moves closer to the control points, ultimately becoming a piecewise linear function (straight lines between the control points) for order = 2. Control points can be repeated and so can knot values. Although the knot values can be any values, not just 0 to 1, usually the knot vector consists of integers starting with 0. If the knots are uniformly spaced, then it is termed a uniform B-spline; if not, it is a non-uniform B-spline. In general, B-spline curves do not pass through any of their control points. However, they can be formulated to pass through the first and last control points, similar to Bzier curves. This requires that the first and last knot values be repeated m times. Examine the B-spline basis to verify this. If k is the number of knots, v is the number of control points, and X is the maximum knot value, then we can compute: X=(v-1)-m+2 k = 2m + X - 1

Chapter 14. Curves

14.6 Parabolic Blending

264

14.6 Parabolic Blending

Parabolic blending is a curve formulation based solely on points on the curve. Given a set of four points, compute a curve between the middle two points by linearly blending parabolas through the first three points and second three points, respectively. In general, a parabola can be represented in vector-matrix form as: P(s) = [ s2 s 1 ] A where A is a 3 by 3 matrix of coefficients that define the curve.
C(u) u=0 s=S t=0 P (s) P
1

u=1 s=1 t=T Q (t) P

s=0

t=1

Figure 14-17. Parabolic Blending of C(u) from P(s) and Q(t). Given three points defining the parabola such that: P0 = P(0), P1 = P(x), P2 = P(1) we can combine these equations into one matrix equation: P0 0 0 1

P1 = x2 x 1 M ( x ) P2 1 1 1 This equation can be solved for M(x): 0 0 1 M ( x ) = x2 x 1 1 1 1

x 1 1 x P0 1 P 1 = ------------- 1 x 2 1 x 2 P 1 2 x x 2 P2 x x 0 0 P2 P0

Chapter 14. Curves

14.7 Rational Formulations

265

The first parabola, P(s), passes through the first three points, and the second parabola, Q(t) passes through the last three points. We now linearly blend these two parabolas (2nd order curves) using the parameter u to obtain a cubic curve between P1 and P2: C(u) = (1 - u) P(s) u Q(t) Note that parameters s and t are linear functions of u: s = S + u ( 1-S ) and t = T u Combining all this, the equation becomes: 1 (1 S) T ------------------- T + ( 1 S ) S + ----------- ----------- ---------------- - S S T 1 1 T P 0 2 2 T 2 T - P 1 1 2 ( 1 S ) - 2 ( S 1 ) C(u) = u 3 u 2 u 1 -- ---------------------- ------------------- T ----------- 2S ----------- S T1 S T1 2 P2 2 ( 1 S ) ( 1 2S ) P3 S 0 ------------------- ------------------- S S 0 1 0 0
2 2

where S is the value of s on P(s) at P1, and T is the value of t on Q(t) at P2, for u in [0,1]. Values for S and T are determined by considering slope continuity between adjacent blended segments, like C1(u) and C2(u) in the figure above. For S= 0.5 and T = 0.5: 1 1 C(u) = u 3 u 2 u 1 -- 2 2 1 0 3 5 0 1 3 4 1 0 P0 1 1 P1 0 P2 0 P 3

This is a simple cubic interpolation scheme given points that must lie on the curve. Each set of four points corresponds to one parabolically blended curve, as illustrated in Figure 1418.

14.7 Rational Formulations

As presented, parametric curves based on polynomials cannot exactly represent the conic

Chapter 14. Curves

14.7 Rational Formulations

266

P (s)

C (u)
1

R (t)

C (u) P
0

Q (t)

Figure 14-18. Parabolic Blending of Spline Curves. forms (circular arcs, etc.). For our quarter circle example, Figure 14-19 shows the error in the circle radius (d = 1 - x2 - y2) at the x and y coordinates generated by P(t) = T A . Note that the error is zero only at t=0, 1 , 2 and 1, the interpolated points. -- -- 3 3

Error

0.0

1 -3

parameter t

2 -3

1.0

Figure 14-19. Error Between Exact Circle Radius and Polynomial Interpolation. To provide greater flexibility, the curve representations can be expanded using homogenous coordinates. wx( t ) = a0 + a1 t + a2 t2 + a3 t3 + . . . + an t n wy( t ) = b0 + b1 t + b2 t2 + b3 t3 + . . . + bn t n wz( t ) = c0 + c1 t + c2 t2 + c3 t3 + . . . + cn t n w( t ) = d0 + d1 t + d2 t2 + d3 t3 + . . . + cn t n In vector form: P( t ) = [ wx( t ) wy( t ) wz( t ) w( t ) ]. Notice that the 3D coordinates of a point now are the ratio of two polynomials and hence

Chapter 14. Curves

14.7 Rational Formulations

267

the term rational curves: wx ( t ) x ( t ) = ------------w( t) wy ( t ) y ( t ) = ------------w(t) wz ( t ) z ( t ) = ------------w(t)

As an example, recall the parameterization of a circle given earlier: 0 R 1 wx wy w = 1 t t 2 2R 0 0 0 R 1 The quadratic (n=2) Bzier formulation yields: 1 0 0 b0 P ( t ) = 1 t t2 2 2 0 b1 1 2 1 b 2 Equating the matrices as before and solving for the control points: b0 1 0 0 1 A = b1 = M be 2 2 0 1 2 1 b2
1

1 0 0 0 R 1 0 R 1 0 R 1 1 2R 0 0 = 1 -- 0 2R 0 0 = R R 1 2 0 R 1 0 R 1 2R 0 2 1 1 1

Compare these results with the previous cubic Bzier control points: Cubic Non-Rational Bzier (Approx.)
b 3 = (0,1)
x

Quadratic Rational Bzier (exact)

b 0 = (0,R,1)
x

b 2 = (0.535,1.015) b 1 = (1.015,0.535)

b 1 = (R,R,1)

b 0 = (1,0)

b 2 = (2R,0,2)

Figure 14-20. Comparison of Non-Rational and Rational Circle Forms.

Chapter 14. Curves

14.8 Review Questions

268

14.8 Review Questions

1. Draw the Bzier control polygons and state the degree for each of the following curves:

2 1 3

The vector form of the cubic Bzier curve (n=3) is: B3( t ) = b0 (1-t)3 + 3 b1 t (1-t)2 + 3 b2 t2 (1-t) + b3 t3 After expanding the coefficients, the matrix form of this cubic equation is:
B ( t) = [ t
3 3

t 1 ] -1 3 -3 1 3 -6 3 0 -3 3 0 0 1 0 0 0

b0 b1 b2 b3

= [T ][Mb][b ]

For the quadratic (n=2) Bzier curve, find: (1) the vector form, and (2) the matrix form, written similar to the examples above. (Recall that 0! = 1.)

Chapter 14. Curves

15.1 Lofted Surfaces

269

Chapter 15. Surfaces

The parametric forms of curves can be extended to surfaces by adding another parametric dimension, from P(t) to P(s, t). The objective is to define a complete surface in terms of certain discrete entities which provide some geometric insight as to the resulting shape. As before with the curve segments, we are considering only a finite portion of an infinite surface. This is generally referred to as a surface patch.

15.1 Lofted Surfaces

The easiest approach is also the oldest: a technique called lofting. A linear blend is made between two curves (typically each is in a parallel plane). The two base curves use a common parameter 't' and range (typically [0,1]), and an additional parameter 's' is introduced to blend these two curves linearly. The ranges of the two parameters, s and t, form a two dimensional space termed the parametric domain of the surface. Given the two base curves: P1(t) and P2(t), the lofted surface is defined as: P( s, t ) = (1- s) P1 (t) + s P2 (t ) parameter domain: 0,1 t 1,1 0 <= s,t <= 1 P10 P2(t) s P11 s 1,0 P00 P1(t) P01 Figure 15-1. Lofted Surface as a Linear Blend of Curves. This type of surface is also called a ruled surface. Lofting is adequate for certain applications (e.g. airfoils) but tends to leave the surface somewhat flat in one direction (Figure

0,0

Chapter 15. Surfaces

15.2 Linear Interior Blending

270

15-2). P2 (t) P1 (t) S2 = blending of P2 and P3

P3 (t)

S1 = blending of P1 and P2 Figure 15-2. Slope Discontinuity at Interior Curves of Lofted Surfaces.

15.2 Linear Interior Blending

One might intuitively feel that any four space curves with four common corner points could be interpolated to yield a surface definition. The interpolation is done with linear blending functions, such as the linear functions B0(x) and B1(x) shown in Figure 15-3. It is essential that the

1 B0(x) = 1 - x

1 B1(x) = 1 - B0(x) = x

Figure 15-3. Linear Blending Functions. blending functions sum to unity for the surface to contain its corner points. The surface is now a linear blend of the curves and the corner points:

Chapter 15. Surfaces

15.2 Linear Interior Blending

271

P( s, t ) =

P1( t ) B0( s )

+ P4( s ) B0( t )

+ P3( t ) B1( s ) + P2( s ) B1( t ) - P00 B0( t ) B0( s ) - P01 B1( t ) B0( s ) - P11 B1( t ) B1( s ) - P10 B0( t ) B1( s ) where: P1 , P2 , P3 , P4 are bounding curves, P00, P01, P11, P10 are the corner vertices. See Figure 15-4. P10 P4(s) P3(t)

P11
s

P2(s) P00
t

P1(t)

P01

Figure 15-4. Surface with Linear Interior Blending of Four Curves. Note the boundary conditions for the boundary curves, i.e. when s and t are 0 and 1: P1( 0 ) = P4( 0 ) = P00 P1( 1 ) = P2( 0 ) = P01 P2( 1 ) = P3( 1 ) = P11 P3( 0 ) = P4( 1 ) = P10

Chapter 15. Surfaces

15.2 Linear Interior Blending

272

Now check the s = t = 0 boundary condition for the surface: P00 1 P00 1 P( s=0, t=0 ) = P1( 0 ) B0( 0 ) + P4( 0 ) B0( 0) P10 P01 0 0 + P3( t ) B1( s ) + P2( 0 ) B1( 0 ) 1 1 - P00 B0( 0 ) B0( 0 ) 1 0 - P01 B1( 0 ) B0( 0 ) 0 0 - P11 B1( 0 ) B1( 0 ) 1 0 - P10 B0( 0 ) B1( 0 ) P( 0, 0 ) = P00 + P00 - P00 = P00 as expected. Note that each corner vertex belongs to two boundary curves and will therefore be added twice when s and t are 0 or 1. This is the reason for the last four terms involving the corner vertices in the P(s,t) equation. As a second check, evaluate P( s, 0 ): P00 1 P( s, t=0 ) = P1( 0 ) B0( s ) + P4( s ) B0( 0 ) P10 0 + P3( 0 ) B1( s ) + P2( s ) B1( 0 ) 1 - P00 B0( 0 ) B0( s ) 0 - P01 B1( 0 ) B0( s ) 0 - P11 B1( 0 ) B1( s ) 1 - P10 B0( 0 ) B1( s ) P( s, t=0 ) = P00 B0( s ) + P4( s ) + P10 B1( s ) - P00 B0( s ) - P10 B1( s ) = P4( s ) as expected Any four space curves (parameterized from 0 to 1) may be used to construct a surface. If P2( s ) and P4( s ) are linear, the equations reduce to a ruled surface.

Chapter 15. Surfaces

15.3 B zier Surfaces

273

15.3 Bzier Surfaces

The Bzier surface formulation is a Cartesian product surface. The general n by m surface is formulated as follows:
n m

P(s, t) =

Bn, i(s)Bm, j(t)bij

i = 0j = 0

where B n, i ( x ) is the Bernstein basis function defined in the previous chapter, n is the degree of the surface in the s direction, and m is the degree of the surface in the t direction. The bij are (n + 1) * (m + 1) control points that form a control net. (Figure 15-5)
b
03

b b
13

s b
01

b b
11

b
22

b b

b
10

b
31

b
33

b
00

b
30

Figure 15-5. Cubic Bzier Surface Control Net. The bi-cubic surface is shown below in blending function format. ( 1 s )3
2 P ( s, t ) = ( 1 t ) 3, 3t ( 1 t ) 2, 3t 2 ( 1 t ), t 3 b 3s ( 1 s ) 3s 2 ( 1 s )

s3 The parameter functions can be factored into two matrices as before:

T P ( s, t ) = TM be b M be S T

Chapter 15. Surfaces

15.4 B-Spline Surfaces

274

where [ b] is a tensor (matrix of vectors) that contains the points on the control net, and T and S are again polynomial basis vectors and Mbe is the Bzier basis change matrix as discussed in the previous chapter. Like the Bzier curve, the Bzier surface closely mimics its control net. The surface contains only the corner points and the three control points nearest each vertex control the surface tangents. Parametric surface tangents are computed by partial derivatives in the s and t parametric directions: P ( s, t ) = P s ( s, t ) s where, P s ( s, t ) = x s ( s, t ) y s ( s, t ) z s ( s, t ) = x ( s, t ) y ( s, t ) z ( s, t ) s s s Given parametric tangents, we compute geometric tangents using ratios: y s ( s, t ) dy ----- ( s, t ) = ---------------x s ( s, t ) dx z s ( s, t ) dz ----- ( s, t ) = ---------------x s ( s, t ) dx z s ( s, t ) dz ----- ( s, t ) = ---------------y s ( s, t ) dy P ( s, t ) = P t ( s, t ) t

Note that because they are ratios of the parametric tangents, the geometric tangents have an extra degree of freedom (a scale factor). Surface normals can also be computed for any parametric point on a surface. The normal is defined by the cross product of the parametric tangent vectors: N ( s, t ) = P s ( s, t ) P t ( s, t ) Surface normals are important for a number of computational geometry applications, including machining of surfaces (locating the position of a cutting tool involves moving along the normal by the radius of the cutter), realistic rendering (lighting model calculations), and finding offset surfaces (the surface which is everywhere equidistant from another surface).

15.4 B-Spline Surfaces

The B-spline curves also generalize to surfaces. Simply use Mbs in place of the Mbe in the surface equations and fill Bx, By, and Bz with B-spline control net coordinates.

Chapter 15. Surfaces

15.4 B-Spline Surfaces

275

x( s, t ) = T Mbs Bx MbsT ST y( s, t ) = T Mbs By MbsT ST z( s, t ) = T Mbs Bz MbsT ST

Chapter 15. Surfaces

16.1 Historical View of Geometric Modeling

276

Chapter 16. Geometric Modeling

The representation and manipulation of geometry is a key ingredient in nearly all computer graphics applications. This is also true in the fields of computer-aided design (CAD) and computeraided engineering (CAE), in which systems strive to create software prototypes of physical objects on which mathematical simulations can be performed to predict the behavior of the objects in real life situations. The current and potential benefits of CAD and CAE systems are improvements in the design and manufacture of new products, where what if questions can be answered by computer simulations instead of building and testing costly, time-consuming prototypes. The technical foundation for CAD systems is the ability to represent the geometry of the parts and assemblies being designed in sufficient detail to allow the computer to assist in design, analysis and manufacture. This field, geometric modeling, has grown with the fields of computer graphics and CAD over the years. A brief chronology of this development is given next.

16.1 Historical View of Geometric Modeling

1950's Computer graphics first appears as a tool for design and manufacturing in large aircraft and auto industries. Early research shows great potential. 1960's Computer graphics hardware is very costly and the software development effort is enormous. Only large companies can justify the investment. ~1963 Sutherland's Sketchpad demonstrates utility of interactive computer graphics for engineering problem-solving in general.

Chapter 16. Geometric Modeling

16.1 Historical View of Geometric Modeling

277

late 1970's The appearance of low-cost minicomputers, coupled with new storage tube CRT terminals, spawn 2D turnkey CAD systems development. Small companies can now develop software and deliver a system affordable to a larger segment of industries. The CAD industry is born. ~1973 Braid's Designing with Volumes thesis and his BUILD system demonstrate the potential for solid modeling in CAD. Also, the PADL research project at Rochester establishes the viability of solid modeling. 1970's Turnkey CAD systems flourish, and add 3D capabilities. A technical debate begins between the drafting versus solid modeling paradigms for CAD. Solid modeling is computationally expensive, making it impractical for most minicomputer systems in place in industry. early 1980's New powerful 32-bit midi -computers enter the marketplace. Turnkey systems struggle to re-write their systems to exploit the power of these computers. The viability of solid modeling increases, encouraging further development by CAD vendors. Solid modeling also gains support as applications appear in computer-aided manufacturing, CAM, such as computer automated machining planning for numerical control machining (NC). Research efforts at universities and industries in CAD grow considerably. mid-1980's Personal computers revolutionize the computer marketplace. PC CAD quickly takes over marketplace, redefining the cost structures and options available to customers, and opening the CAD market to all industries, large and small. The established turnkey market stagnates. Solid modeling has become a part of every large CAD system.

Chapter 16. Geometric Modeling

16.2 CAD and Geometric Modeling

278

1990s + Parametric and constraint-based design systems gain attention. Integration of design and manufacturing becomes more important.

16.2 CAD and Geometric Modeling

CAD systems share an obsession with geometry. As CAD systems evolved, so did the treatment of geometry:

2D wireframe models geometry consists of lines and curves in 2D: (x,y), objects are created using drafting methods, 3D is simulated using multiple views. 3D wireframe models like 2D wireframe data, but 3D: (x,y,z), space curves now possible. surface models surfaces with boundary curves between points, skin of aircraft, auto, NC possible with manual guidance. solid models oriented surfaces + topology, robust shape description, complete spatial enumeration, automation (shape understanding) is aided greatly. features models are an extension of solid models to include product data in a form that supports integration among design, analysis and manufacturing programs.

Chapter 16. Geometric Modeling

16.3 Solid Geometric Modeling

279

16.3 Solid Geometric Modeling

A serious limitation with wireframe geometry for modeling real objects is that some necessary information is missing. Wireframe and surface representations do not completely and unambiguously define the total geometry. These systems require that the spatial occupancy of the object be interpreted by a human. Consider the information necessary to compute the volume of an object stored as a set of polygons. Do the structures discussed so far contain sufficient information? Generally not. Another problem is that it is possible to accidentally design objects that are physically impossible to build, due to dangling edges and faces, and faces or edges that pass through themselves. To have some assurance that the resulting geometry represents an object that is physically realizable, a more rigorous way to construct and store 3D geometry is necessary. At the same time, there is interest in developing computer-aided design systems with capabilities for designing in three dimensions with more powerful construction methods than simple points, lines and surfaces. Solid modeling began in the early 1970's at universities in England and the U.S. The primary goal for solid modeling was, and still is, to develop a complete and unambiguous computer representation for physically realizable objects that can be used for all engineering design, analysis and manufacturing purposes with a high degree of automation. Solid modeling is key to computeraided design (CAD), computer-aided manufacturing (CAM) and automation. A simple test can be made to assess the robustness and completeness of a geometric model: Is the point (x,y,z) inside, outside or on the object? Solid modeling systems can answer this question without human intervention. Can wireframe systems (for example wireframe or surface models)? No.

16.4 Solid Modeling Methods

No solid modeling system today is able to model all manufactured parts exactly. Each system is limited in some way; generally due to its inability to deal with complex geometry, such as curved surfaces. In these cases the system either gives up ( I can't do that ) or makes some approximation and continues.

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

280

In general, solid modeling methods vary according to how: 1. the model is constructed: languages and/or graphical interaction, set operations, sweeps, Euler operations 2. the model is stored (i.e. the internal data structure): CSG BREP Hybrid (combination of CSG and BREP) We will examine two of the fundamental approaches: constructive solid geometry (CSG) and boundary representation (BREP).

16.4.1 Boundary Representation

Probably the most popular method for storing geometric models is boundary representation, or BREP. As the name implies, BREP models maintain a complete description of all the surfaces bounding the object. A typical BREP stores all the faces (surfaces), edges and vertices on the object in a richly linked data structure. This structure stores: 1. geometry: surface equations, points, curve equations. 2. topology: how the geometry is connected. vertices, edges, loops, faces, shells, objects. The typical structure is called a face-edge-vertex graph, or FEV. The best description of this structure is everything points to everything else.

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

281

Object

Face

Loop

Vertex Edge Vertex Edge Edge

Figure 16-1. Illustration of a BREP Data Structure.

16.4.2 Constructive Solid Geometry (CSG)

CSG, or primitive instancing, has a natural appeal to combining objects by applying Boolean logic to volumes. For example, making a hole, slot, pocket, etc. is made much simpler with CSG operations because one simply instructs the system to subtract one object from another. CSG also strongly relates to manufacturing operations, another reason for its popularity in CAD and CAM. Some engineering applications relate to CSG construction methods somewhat naturally, for example: 1. creating a mold for a given part using CSG modeling by subtracting the part from a block. 2. simulating machining operations by subtracting the cavities made by the tools from the part. 3. checking for interference between parts by computing their Boolean intersection.

Intuitively, one realizes that these operations produce valid solids (except for some surfaceon-surface situations). Although different modelers use different sets of primitives, the ones

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

282

generally common to all are the Sphere and Cone, etc. as illustrated in the figures.

object = Sphere(P, radius)

Figure 16-2. Sphere Primitive and Functional Notation.

object = Cone(P, axis, radius1, radius2)

Figure 16-3. Cone Primitive and Functional Notation. 1 Swept Solids, also called 2 -- D solids: 2 Planar profile projected by extrusion: Slab.

Figure 16-4. Extruded or 2-1/2D Primitive. Planar profile revolved about an axis: Surface of revolution.

Figure 16-5. Surface of Revolution Primitive. Objects can be stored in CSG trees, also. A CSG tree is a binary tree data structure with Boolean operations at its branches and primitives at the leaves. Stored CSG models are in fact only a definition of how to make the model from primitives. This is called an unevaluated form, because each application that wants to use the

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

283

- -

Figure 16-6. CSG Tree Representation (Right) of an Object (Left). model must evaluate it for its own purposes. One of the methods used to evaluate a CSG model for realistic display is ray casting as discussed earlier.

16.4.3 Evaluating CSG Objects: A 2D Example

We can describe enclosed 2D areas with expressions, where the operators are '+' for Boolean addition, '-' for subtraction and '*' for intersection, and the operands are primitive objects, or primitives, such as CIRCLE(Xc, Yc, Radius) and BOX(Xc, Yc, Size). For example, the expression: OBJECT = BOX1(0, 0, 1) + CIRCLE(1, 0, 1) - BOX2(0, 0, 0.5) creates the OBJECT shown in Figure 16-7. Letting the operator P mean primitive, OBJECT could be stored in a binary tree data structure, a CSG tree, shown in Figure 16-8. We can perform computations on a CSG tree to determine if a given point (x,y) is inside or outside the object described by the tree. For our example, let the value TRUE mean inside or on the object, and FALSE mean outside. We will design a function TestPoint(x,y,tree) which will return TRUE (actually the value 1) or FALSE (the value 0) for the CSG tree pointed to by tree. First define two data structures, CSGTree and Primitive, as illustrated in Figure 16-9. C code for the CSGTree structure could be:

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

284

BOX1

Y CIRCLE

BOX2 Figure 16-7. Example Object (right) formed by Primitives (left).

OBJECT

BOX 0,0,0.5

BOX 0,0,1

CIRCLE 0,0,1

Figure 16-8. CSG Tree Representation of the Object in Figure 16-7.

CSGTree +,-,* left P prim * Primitive right

Figure 16-9

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

285

typedef struct CSGTree { int operator; struct CSGTree *left, *right; struct Primitive *prim; } CSGTree; and Primitive could be any appropriate data representation for storing the primitives data. Define the values for the operator field: ADD, SUBTRACT, INTERSECT and PRIMITIVE. The fields left and right point to sub-trees for the +, - and * binary operators. They are not used for the P operator. The field prim points to a structure of type Primitive only if the operator is P. The function to evaluate if a given point is inside or on a CSG object is coded (in C) using recursion as shown in Figure 16-10.
Boolean TestPoint( x, y, tree ) float x, y; CSGTree *tree; { switch( tree->operator ) { case ADD: return( TestPoint(x,y,tree->left) || TestPoint(x,y,tree->right) ); case SUBTRACT: return( TestPoint(x,y,tree->left) && ! TestPoint(x,y,tree->right) ); case INTERSECT: return( TestPoint(x,y,tree->left) && TestPoint(x,y,tree->right) ); case PRIMITIVE: if( tree->prim->type == CIRCLE ) return( TestCircle(x,y, tree->prim) ); if( tree->prim->type == BOX ) return( TestBox(x,y, tree->prim) ); } }

Figure 16-10. CSG Evaluation Function TestPoint. The primitive routines TestCircle and TestBox must return TRUE if the given point is inside the particular instance of the primitive, or FALSE otherwise. All calls to TestPoint end up in calls to primitive routines. Adding new primitives is simply a matter of coding their TestXXX routines.

Chapter 16. Geometric Modeling

16.4 Solid Modeling Methods

286

For some applications, such as ray casting CSG trees, two-state logic is inadequate because it is necessary to know when a point is ON the object as well. Figure 16-11 gives the three-state logic tables for the Boolean operators.
B ON IN ON ON

UNION IN A ON OUT

IN IN IN IN

OUT IN ON OUT

DIFFERENCE IN A ON OUT

IN OUT OUT OUT

B ON ON ON OUT B

OUT IN ON OUT

INTERSECTION: IN A ON

IN IN ON

ON ON ON OUT

OUT OUT OUT OUT

OUT OUT

Figure 16-11. Three-State CSG Truth Tables.

Chapter 16. Geometric Modeling

Index

I-1

Index
Numerics
2D scaling 1-80 3D endpoint code 1-154 3D pipeline 1-150

A
achromatic 1-21 additive primaries 1-25 AIM-based viewing 1-164 algebraic curves 1-254 aliasing 1-247 ambient light 1-219 ambiguous picks 1-98 Apple Macintosh Toolbox 1-86 application context 1-98 ASCII 1-50 aspect ratio 1-72 azimuth 1-147

B
back-facing 1-198 basic definition 1-119 beam movement 1-65 Bernstein basis functions 1-259 Bezier curve definition 1-259 binomial coefficient 1-259 bitmap display 1-15 blending functions 1-270 boundary representation 1-280 bounding rectangle 1-102 Bresenham 1-44 bump mapping 1-248 byte 1-50

color coordinates 1-27 color lookup table (CLUT) 1-23 color model 1-27 color space 1-27 colorfulness 1-33 compacted sequential list 1-107 composite transformation matrix 1-83 computational geometry 1-170 concatenation 1-82 concave object 1-198 constructive solid geometry 1-280 control net 1-273 control points 1-259 control polygon 1-259 convex object 1-198 CPU 1-67 CPU memory 1-67 cross hairs 1-94 cross product 1-175 CSG 1-280 CSG tree 1-282 CSL 1-107 CTM 1-124 CTM stack 1-125 cubic polynomials 1-255 current transformation matrix 1-124 cursor 1-93

D
decay 1-13 depth comparisons 1-199 device coordinates 1-70 dialog windows 1-89 diffuse reflectance 1-216, 1-246, 1-247, 1-248 direct color 1-23 direction vector 1-170 display instruction 1-61 display memory 1-63 display programming 1-61 distortion 1-72 divide and conquer 1-205 dot product 1-174 dots per inch (dpi) 1-10 DPU 1-61 DPU memory 1-67 drawing attributes 1-91 drawing mode 1-91 drawing primitives 1-88 drawing state 1-91 drum plotter 1-4 ducks 1-253 DX vector 1-232 dynamic display manipulation 1-67

C
center of projection 1-135 character definition grid 1-49 character font 1-49 character spacing 1-51 chroma 1-33 chromatic 1-21 chromaticity 1-31 chromaticity diagram 1-31 clipping coordinate system 1-158 clipping coordinates 1-158 clipping plane coefficient vectors 1-155 coefficient matrix 1-255 color components 1-22

Index

I-2

E
edge-driven 1-205 elevation 1-148 event 1-88, 1-90 event loop 1-90 extent 1-72 extrusion 1-282 eye coordinate system 1-135 eye coordinates 1-144 EYE-based viewing 1-166

inside/outside test 1-233 instance 1-119 instance table 1-119 instance transformations 1-123, 1-189 intensity depth cueing 1-132 intensity interpolation 1-223 interlace 1-14 interpolation 1-255 intersection of half-spaces 1-234 intersection tree 1-230 inversion 1-253

F
facets 1-223 far clipping plane 1-149 feedback 1-95 field-of-view 1-151, 1-163 filling 1-56, 1-209 film recorders 1-10 fixed width spacing 1-51 flat shading 1-223 flatbed plotter 1-5 flicker 1-14 font 1-51 frame buffer 1-15

K
kernel-based 1-86 kinetic depth 1-132 knot 1-256 knot vector 1-263

L
laser printers 1-10 lighting model 1-215 line of sight 1-197, 1-199 linear blend 1-269 linked lists 1-111 list compaction 1-109 local control 1-263 local intelligence 1-4 local refresh 1-63 lofting 1-269 luminance 1-21

G
geometric modeling 1-170 geometric tangents 1-274 glyph 1-49 Gouraud shading 1-223 graphical user interface 1-93 gray levels 1-17 gray scale display 1-17 growing lines 1-205 GUI 1-93

M
mach bands 1-224 mapped color 1-23 matrix equation 1-78 Microsoft Windows 1-87 modes 1-104

H
Hermitian interpolation 1-257 hierarchical scene 1-121 highlight 1-98, 1-105 HLR 1-130 homogeneous coordinates 1-142 homogeneous dot product 1-154 homogeneous transformation matrix 1-140 HSR 1-130

N
normalization transform 1-158 near clipping plane 1-149 network-based 1-86 non-uniform B-spline 1-263 normalized homogeneous coordinates 1-142 noun-verb paradigm 1-104 NTSC 1-29

I
identity transformations 1-82 illumination model 1-215 image plane 1-135 implicit representation 1-252 implicitization 1-253

O
object table 1-118 Off-line 1-6 on the fly execution 1-63 on-line 1-7

Index

I-3

ordinary coordinates 1-142 orthographic projection 1-131 outline fonts 1-50

P
perspective matrix 1-143 page plotting system 1-55 palette 1-24 parallel light source 1-221 parametric domain 1-269 parametric knots 1-262 parametric polynomial 1-254 parametric representation 1-252 parametric tangents 1-256, 1-274 pen code 1-43 pen plotters 1-4 persistence 1-13 perspective depth cueing 1-131 perspective distance (d) 1-136 perspective factor 1-137 Phong shading 1-225 picking 1-97 pictorial renderings 1-130 picture plane 1-135 pixel 1-14 pixel coordinates 1-55 pixel memory 1-15 pixel patterns 1-56 pixel transfer functions 1-57 pixel value 1-15 plane coefficient vector 1-154 plane equation 1-154 plotting package 1-42 point containment test 1-233 point light source 1-221 point plotting display 1-13 points at infinity 1-143 Polygon clipping 1-206 polynomial basis 1-254 polynomial forms 1-254 poor man's hidden line removal 1-197 position vector 1-170 positioning constraints 1-94 primary colors 1-22 primitive instancing 1-281 principle of least astonishment 1-93 procedural texture mapping 1-249 projection axonometric 1-135 cabinet 1-135 cavalier 1-135 isometric 1-135 oblique parallel 1-135

orthographic parallel 1-135 parallel planar 1-135 perspective planar 1-135 projection plane 1-135 projector 1-135 proportional spacing 1-51

Q
quick-rejection 1-234

R
random vector 1-53 raster 1-53 raster display 1-14 raster display processor 1-15 raster location 1-54 raster unit 1-53, 1-54 rasterization 1-53 rational curves 1-267 reflectance model 1-215 refraction 1-230 refresh 1-13 refresh burden 1-63 region code 1-73 repaint 1-68 resolution 1-6 Robert's HLR algorithm 1-200 rubber-band line 1-95 ruled surface 1-269

S
S vector 1-232 scalar product 1-174 scalar triple product 1-177 scan-cycle 1-14 scan-line 1-14 scan-line coherence 1-205 scan-plane 1-204 screen context 1-98 screen integrity 1-111 segments 1-86 selection 1-105 selection region 1-103 shading model 1-215 shape design system 1-254 SKETCHPAD 1-1 slope continuity 1-257 spatial occupancy 1-279 spectral distribution 1-20 spectral reflectance curve 1-20 specular reflection 1-217 spline 1-253 spline curve 1-256

Index

I-4

Spooling 1-7 spooling 1-7 spotlight source 1-221 stack pointer 1-125 step 1-4 stepping algorithm 1-44 stepping motor 1-6 step-size 1-6 stereoscopic views 1-132 strip-page plotting 1-55 strokes 1-50 subtractive primaries 1-26 surface normals 1-274 surface of revolution 1-282 surface patch 1-269 surface tangents 1-274 swept solids 1-282

magnitude 1-172 plane equation 1-179 unit vector 1-174 vector display processor 1-61 vector generator 1-14 vector product 1-175 vector volume 1-177 verb-noun paradigm 1-104 video 1-14 video lookup table (VLT) 1-23 view vector 1-144 viewing frustum 1-151, 1-164 viewing sphere 1-148 viewing volume 1-151 viewport 1-70

W
window 1-86 window coordinates 1-91 window systems 1-88 world coordinates 1-70

T
tesselation 1-223 texels 1-246 texture 1-246 texture coordinates 1-246 texture mapping 1-246 texture space 1-246 tile systems 1-88 topology 1-280 transformation 1-77 2D rotation 1-81 2D translation 1-80 3D rotation 1-140 3D scaling 1-140 3D translation 1-140 concatenation 1-82 trichroma 1-22 trichromatic vector 1-220 tri-stimulus theory 1-22 two dimensional clipping 1-73

X
X Window System 1-87

Z
z-clipping 1-149

U
unevaluated form 1-282 uniform B-spline 1-263 uniform scaling 1-72 up direction 1-188 update 1-64, 1-68 user model 1-98

V
viewing transformation 1-144 vanishing point 1-131 variable width spacing 1-51 vector cross product 1-175

Index

APPENDIX. The GRAFIC Package.

Computer Science - Introduction to Problem Solving
No ratings yet
Computer Science - Introduction to Problem Solving
38 pages
Chapter 14 Questions With Answers - Mod 7 Quiz
0% (1)
Chapter 14 Questions With Answers - Mod 7 Quiz
3 pages
Computer Graphics - Opengl
No ratings yet
Computer Graphics - Opengl
150 pages
Computer-Graphics Book 2 PDF
No ratings yet
Computer-Graphics Book 2 PDF
0 pages
Text Book For Computer Graphics
100% (1)
Text Book For Computer Graphics
181 pages
UNIT 1 Introduction To Computer Graphics
No ratings yet
UNIT 1 Introduction To Computer Graphics
182 pages
DS - Chapter 1 and 2
100% (1)
DS - Chapter 1 and 2
24 pages
Chapter 4 - Programming in C
No ratings yet
Chapter 4 - Programming in C
159 pages
Computer Graphics I UNIT Notes
100% (1)
Computer Graphics I UNIT Notes
24 pages
Computer Graphics Notes PDF
No ratings yet
Computer Graphics Notes PDF
192 pages
Date: - 01 - 2021: Final Examination Fall 2020
No ratings yet
Date: - 01 - 2021: Final Examination Fall 2020
2 pages
Hill CH 1 Ed 3
No ratings yet
Hill CH 1 Ed 3
60 pages
Windows Programming
No ratings yet
Windows Programming
27 pages
The Importance of Algorithms
No ratings yet
The Importance of Algorithms
22 pages
Line Scan Conversion in Computer Graphics
No ratings yet
Line Scan Conversion in Computer Graphics
48 pages
Features 1 B
No ratings yet
Features 1 B
94 pages
Networking Essentials 1206 Bac 2005230 Joy Ntinyari Kimathi
No ratings yet
Networking Essentials 1206 Bac 2005230 Joy Ntinyari Kimathi
5 pages
4 Assembly Language Program Example
No ratings yet
4 Assembly Language Program Example
13 pages
Course Log - Theory of Programming Languages
No ratings yet
Course Log - Theory of Programming Languages
6 pages
CNE Tutorial 02 - Real - LAN and LAN With Packet Tracer
No ratings yet
CNE Tutorial 02 - Real - LAN and LAN With Packet Tracer
18 pages
First Year
No ratings yet
First Year
21 pages
Unit 2 Ooad Question
No ratings yet
Unit 2 Ooad Question
2 pages
Computer Architecture
No ratings yet
Computer Architecture
10 pages
3 - C++ Programming
No ratings yet
3 - C++ Programming
44 pages
Computer Graphics Notes - Unknown
No ratings yet
Computer Graphics Notes - Unknown
110 pages
Burrows-Wheeler Transform - Wikipedia, The Free Encyclopedia
No ratings yet
Burrows-Wheeler Transform - Wikipedia, The Free Encyclopedia
10 pages
Ccs0003 Computer Programming 1 Lec Syllabus
No ratings yet
Ccs0003 Computer Programming 1 Lec Syllabus
6 pages
Modeling and Simulation Lab 02
No ratings yet
Modeling and Simulation Lab 02
7 pages
PIAIC AIoT Q1 Syllabus Final
No ratings yet
PIAIC AIoT Q1 Syllabus Final
3 pages
Lecture 2 - Programming With VB
No ratings yet
Lecture 2 - Programming With VB
17 pages
CG Book 3rd Edition
No ratings yet
CG Book 3rd Edition
385 pages
B-Spline Surface Reconstruction by Control Point Optimization Using Genetic Algorithm-135
No ratings yet
B-Spline Surface Reconstruction by Control Point Optimization Using Genetic Algorithm-135
9 pages
Computer Organization & Architecture
No ratings yet
Computer Organization & Architecture
55 pages
CH 9 Planning and Costing
No ratings yet
CH 9 Planning and Costing
30 pages
Surface Science Surface and Adsorbate Structure V5 KI
No ratings yet
Surface Science Surface and Adsorbate Structure V5 KI
67 pages
It - Com 424
No ratings yet
It - Com 424
35 pages
Lecture Notes 1.3.4
No ratings yet
Lecture Notes 1.3.4
4 pages
C Graphics
No ratings yet
C Graphics
31 pages
Dikshant International School: Project On Computer Networking
No ratings yet
Dikshant International School: Project On Computer Networking
17 pages
Principal of Programming Language
No ratings yet
Principal of Programming Language
67 pages
C Programming
No ratings yet
C Programming
205 pages
Advanced Animation and Rendering Techniques
No ratings yet
Advanced Animation and Rendering Techniques
5 pages
Computer Graphics
No ratings yet
Computer Graphics
10 pages
Data Structure Lab - MCA16-L
No ratings yet
Data Structure Lab - MCA16-L
57 pages
Automata and Compiler Design - Lecture Notes On UNIT 1
No ratings yet
Automata and Compiler Design - Lecture Notes On UNIT 1
25 pages
Lab3 WindowForms EN
No ratings yet
Lab3 WindowForms EN
18 pages
Lab 5 - Combinational Logic-Adders and Subtractors
No ratings yet
Lab 5 - Combinational Logic-Adders and Subtractors
11 pages
Computer Graphics & Image Processing
No ratings yet
Computer Graphics & Image Processing
117 pages
تحليل عددي الجزء الثاني PDF
No ratings yet
تحليل عددي الجزء الثاني PDF
57 pages
Fortran - Paul Tackley
No ratings yet
Fortran - Paul Tackley
32 pages
CS8451 DESIGN AND ANALYSIS OF ALGORITHMS QUESTION BANK - Watermark PDF
50% (2)
CS8451 DESIGN AND ANALYSIS OF ALGORITHMS QUESTION BANK - Watermark PDF
47 pages
021 Windows Forms Intro
No ratings yet
021 Windows Forms Intro
51 pages
Chapter 5 Application Software: Tools For Productivity: Computers Are Your Future, 10e (Coyle)
No ratings yet
Chapter 5 Application Software: Tools For Productivity: Computers Are Your Future, 10e (Coyle)
9 pages
Computer Graphics and Image Processing
33% (3)
Computer Graphics and Image Processing
14 pages
15cs204j-Algorithm Design and Analysis
No ratings yet
15cs204j-Algorithm Design and Analysis
3 pages
Topological Sorting
No ratings yet
Topological Sorting
8 pages
Introduction To OOAD What Is OOAD? What Is UML? What Are The United Process (UP) Phases
No ratings yet
Introduction To OOAD What Is OOAD? What Is UML? What Are The United Process (UP) Phases
41 pages
Advanced Unix Programming
From Everand
Advanced Unix Programming
Prof. N. B Venkateswarlu
No ratings yet
C Programming: Core Concepts and Techniques
From Everand
C Programming: Core Concepts and Techniques
William Smith
No ratings yet
Mobile Python: Rapid prototyping of applications on the mobile platform
From Everand
Mobile Python: Rapid prototyping of applications on the mobile platform
Jürgen Scheible
4.5/5 (2)
Equity of Cybersecurity in the Education System: High Schools, Undergraduate, Graduate and Post-Graduate Studies.
From Everand
Equity of Cybersecurity in the Education System: High Schools, Undergraduate, Graduate and Post-Graduate Studies.
Joseph O. Esin
No ratings yet
Worldskills Mechanical Engineering
No ratings yet
Worldskills Mechanical Engineering
8 pages
Quick Start Guide Xerox 7142
0% (1)
Quick Start Guide Xerox 7142
32 pages
Subject: Concepts in Computer Engineering (BECME303T) : Unit I Input/Output Units and Computer Memory
No ratings yet
Subject: Concepts in Computer Engineering (BECME303T) : Unit I Input/Output Units and Computer Memory
69 pages
Computer System Components
No ratings yet
Computer System Components
47 pages
PrinterDriver Operation Manual
No ratings yet
PrinterDriver Operation Manual
36 pages
1stpucchapter 2
No ratings yet
1stpucchapter 2
25 pages
FORM 3 Output Devices Test
No ratings yet
FORM 3 Output Devices Test
2 pages
Lecture 3 Computer Fundamental
No ratings yet
Lecture 3 Computer Fundamental
7 pages
Office
100% (1)
Office
24 pages
BBA-104 Unit-1 Fundamentals-of-Computers Notes
No ratings yet
BBA-104 Unit-1 Fundamentals-of-Computers Notes
24 pages
CGMT PPT Unit-1
No ratings yet
CGMT PPT Unit-1
81 pages
ValueCutRegistratonGuide SAi
No ratings yet
ValueCutRegistratonGuide SAi
5 pages
MCQ Cad Cam Cae Smshaikh
100% (2)
MCQ Cad Cam Cae Smshaikh
10 pages
Output Devices Notes
No ratings yet
Output Devices Notes
2 pages
22052-Manufacturing Technology - Ii PDF
No ratings yet
22052-Manufacturing Technology - Ii PDF
16 pages
Define Parameter
No ratings yet
Define Parameter
4 pages
US Navy Course - Blueprint Reading and Sketching NAVEDTRA 14040
No ratings yet
US Navy Course - Blueprint Reading and Sketching NAVEDTRA 14040
185 pages
UCEST105 M1 Ktunotes.in
No ratings yet
UCEST105 M1 Ktunotes.in
54 pages
Computer Graphics Questions and Answers
No ratings yet
Computer Graphics Questions and Answers
8 pages
Ict 6
No ratings yet
Ict 6
3 pages
4 KMDF3 Red DIL533 Y516
No ratings yet
4 KMDF3 Red DIL533 Y516
8 pages
Design and Development of A Computer Numeric Controlled 3D Printer, Laser Cutter and 2D Plotter All in One Machine
No ratings yet
Design and Development of A Computer Numeric Controlled 3D Printer, Laser Cutter and 2D Plotter All in One Machine
7 pages
Sublimation Printable White Heat Transfer Vinyl
No ratings yet
Sublimation Printable White Heat Transfer Vinyl
25 pages
Arduino Based Cost Effective CNC Plotter Machine
No ratings yet
Arduino Based Cost Effective CNC Plotter Machine
4 pages
0417 s12 QP 12
No ratings yet
0417 s12 QP 12
12 pages
DXF Aama
50% (2)
DXF Aama
7 pages
MCS 053 Final
No ratings yet
MCS 053 Final
28 pages
QP CSC Q0402 Draughtsman Mechanical 1 02.07.2018
No ratings yet
QP CSC Q0402 Draughtsman Mechanical 1 02.07.2018
34 pages
Summary Notes
No ratings yet
Summary Notes
63 pages