Lesson4 PDF

Preprocessing
n Duration: 3 hr
n Outline:
1. Introduction to Matlab
2. Fundamentals
3. Spatial domain processing
4. Frequency domain processing
Preprocessing
n Duration: 3 hr
n Outline:
2. Fundamentals
Matlab working environment
§ Matlab desktop:
Ø command window
Ø workspace
Ø command history window
Ø current directory window
Ø figure window
Ø editor window
Matlab working environment (cont)
Current directory window
Command window
workspace
Figure window
History command
window
Editor window
Basic commands and formatting
Basic commands and formatting (cont)
Basic commands and formatting (cont)
Conditional statements and loops
§ Conditional statements
Conditional statements and loops (cont)
§ Loops
Preprocessing
n Duration: 3 hr
n Outline:
2. Fundamentals
Digital image representation
§ Monochrome image ßà image: 2D light intensity function f(x,y)

Where (x,y): spatial coordinates
f value: brightness (or gray level) of the image at point (x,y)
origin y
x
Digital image representation
§ Digital image: be discretized in spatial coordinates and

brightness
§ Digital image: matrix, element is called picture element

(pixel)
§ Truecolor image: matrix of pixel value is a 3D array

with size M- by-N-by-3.
Image data
§ Binary: 0 and 1
§ Gray scale: black, shades of gray, white. Typically 256

quantization levels
§ Color: three primary color components, e.g RGB, HSV,

Lab, Luv, CMYK, etc.
§ Resolution: 1024x1024, 512x512, 256x256, 352x240, …

Coordinate conventions
In books In Matlab image processing toolbox

Image as matrix
origin
⎡ f (1,1) f =
f (1,2) ... f (1, N ) ⎤⎥
⎢
⎢ f (2,1) f (2,2) ... f (2, N ) ⎥
f =⎢ ⎥
⎢ ... ⎥
⎢⎣ f (M ,1) f (M ,2) ... f (M , N ) ⎥⎦
Working with images (1/2)
§ Reading an image
>> imread(‘filename’);
e.g.
Working with images (2/2)
§ Displaying an image
>> figure, imshow(im);
>> figure, imshow(im [low high]);

?
?
?
?
Preprocessing
n Duration: 3 hr
n Outline:
2. Fundamentals
Spatial domain processing
§ Intensity transformations (point processing)
§ Histogram processing
§ Spatial filtering
Expression of spatial domain processing
g(x,y) = T[f(x,y)]
output input
operator: defined
over a specified
neighborhood
about point (x,y)
Image f(x,y)
Intensity transformation
§ Operator T: size of neighborhood = 1x1 à point processing

§ Arithmetic operations
§ Negative
§ Contrast stretching
§ Log transformation
§ Power-law transforms
Arithmetic operations: addition, subtraction
Addition
Subtraction
Arithmetic operations: multiplication,
division
Multiplication
Division
Negative
Example of leaf segmentation
Negative
Original mammogram Negative mammogram

Contrast stretching
§ It is normally
controlled by a
piecewise linear
transformation
§ Thresholding:
r1 = r2, s1 = 0, s2
= L-1.
Contrast stretching Thresholding

Log transformation
Original image After processing c = 1

Power-law transformation
§ Plots of the
equation s = c.rγ, for
various values of γ
(c = 1 in all cases)
Power-law transformation
Original γ=3
image c=1
γ=4 γ=5
c=1 c=1
Image histogram
§ Graphical representation of the lightness/color

distribution in a digital image.
§ Number of pixels per each intensity value

Examples of Dark image
image
histograms
Bright image
Low-contrast image
High-contrast image
Histogram equalization
Re-assigning the intensity values of pixels in the input image

such that the output image histogram is flat.
Contrast stretching
Contrast stretching
Some typical lowpass filters
An example of image after lowpass filtering
n2
h= n1
Median filtering
§ Lowpass filters cause blurring, which is critical for

images with sharp edges, but not critical for image of
smooth contrast.
§ To avoid blurring, median filtering is a solution
§ Median filtering is a nonlinear process especially useful

for reducing impulse, or salt-and-pepper noise
Averaging mask vs. median filter
Noisy image After 3x3 After 3x3

averaging masking median filtering
Highpass filters
§ Some highpass spatial filters:
§ Note: if the result is over the range [0, L-1], need to scale
and level-shift to remain in [0, L-1]
High-boost filters
§
é- 1 - 1 - 1ù é- 1 - 1 - 1ù é0 0 0ù
1ê ú 1ê
with A ³ 1, -1 9A -1 -1 = -1 8 - 1 ú + ê0 A -1 0ú
9 ê ú 9ê ú ê ú
êë- 1 - 1 - 1úû êë- 1 - 1 - 1úû êë0 0 0 úû
Original image A = 1.1

High-boost filters
§
é- 1 - 1 - 1ù é- 1 - 1 - 1ù é0 0 0ù
1ê ú 1ê
with A ³ 1, -1 9A -1 -1 = -1 8 - 1 ú + ê0 A -1 0ú
9 ê ú 9ê ú ê ú
êë- 1 - 1 - 1úû êë- 1 - 1 - 1úû êë0 0 0 úû
A = 1.15 A = 1.2
Image after high-boost filtering
Preprocessing
n Duration: 3 hr
n Outline:
2. Fundamentals
Discrete Fourier Transform (DFT)
DFT in image processing
Smoothing an image by low-pass filter
§ Smoothing is achieved by dropping out the high frequency

components using low pass filters
§ Basic model: G(u,v) = H(u,v) F(u,v)
F(u,v): DFT of image, H(u,v): filter transfer function

Ideal low-pass filter
Butterworth low-pass filter
Gaussian low-pass filter
Gaussian low pass filter to connect broken text
Gaussian low pass filter to remove blemishes
Sharpening by high-pass filter
§ Sharpening is to enhance the edges and fine details in an

image.
§ Sharpening is achieved by high pass filters
§ Basic model: reverse of low pass

Matlab
§ imread
§ imshow
§ imfilter
§ conv2
§ medfilt2
§ imnoise
§ fspecial
§ dftfilt
§ freqz
§ hist

Lesson4 PDF

Uploaded by

Copyright:

Available Formats

Lesson4 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lesson4 PDF

Uploaded by

Copyright:

Available Formats

Preprocessing

Current directory window

§ Monochrome image ßà image: 2D light intensity function f(x,y)

§ Digital image: be discretized in spatial coordinates and

§ Digital image: matrix, element is called picture element

§ Truecolor image: matrix of pixel value is a 3D array

§ Gray scale: black, shades of gray, white. Typically 256

§ Color: three primary color components, e.g RGB, HSV,

§ Resolution: 1024x1024, 512x512, 256x256, 352x240, …

In books In Matlab image processing toolbox

>> figure, imshow(im [low high]);

§ Intensity transformations (point processing)

§ Operator T: size of neighborhood = 1x1 à point processing

Original mammogram Negative mammogram

Contrast stretching Thresholding

Original image After processing c = 1

§ Intensity transformations (point processing)

§ Graphical representation of the lightness/color

§ Number of pixels per each intensity value

Re-assigning the intensity values of pixels in the input image

§ Intensity transformations (point processing)

§ Lowpass filters cause blurring, which is critical for

§ To avoid blurring, median filtering is a solution

§ Median filtering is a nonlinear process especially useful

Noisy image After 3x3 After 3x3

Original image A = 1.1

§ Smoothing is achieved by dropping out the high frequency

§ Basic model: G(u,v) = H(u,v) F(u,v)

F(u,v): DFT of image, H(u,v): filter transfer function

§ Sharpening is to enhance the edges and fine details in an

§ Sharpening is achieved by high pass filters

§ Basic model: reverse of low pass

You might also like