INTRODUCTION TO

IMAGE PROCESSING
Bhogendra Mishra PhD
Sept 3, 2023
 PREREQUISITE OF THIS COURSE
• This is a Remote Sensing/Computer Science Course
• It will involve a fair amount of mathematics
• Calculus, linear algebra, geometry,
• Probability
• Analog/ digital signal processing
• Graph theory etc.
• It will involve the modeling and design of real system – one final course project
• Programming skills with Python, R, MATLAB
• jupyter notebook/Google Colab
 OBJECTIVE OF THIS COURSE

This is a graduate-level course

• Research oriented
• Paper reading & presentation
• Final project & presentation
• Review on the state-of-the-art technology
• Understanding – innovation
• Your own innovative and original work/opinion/interpretation
• Basic knowledge  research frontiers
• Learn through reading recent papers
 Final Requirements
REQUIREMENT OF FINAL PROJECT Select a topic and write a one-page proposal
(due Oct 8th)
• Option 1: A complete research project Progress report
 Introduction (problem formulation/definition)
Research work and report writing
 literature review Oral presentation
Section 001: in class presentation
 the proposed method and analysis
Section J60: prerecorded video
 experiment
 conclusion
Final project report
 reference
Written report
Option 2: A survey research
Report format: the same as a conference paper
• A well-defined problem or topic
Executable code must be submitted with clear
• a complete list of previous (typical) work on this
problem (15+ papers under the topic) comments except for a survey study
• clearly and briefly describe the topic
• analyze each method/group and compare them Academic integrity (avoiding plagiarism)
• give the conclusion and list of references don’t copy other person’s work
describe using your own words
Teamwork is acceptable for a research complete citation and acknowledgement whenever
project (Option 1) you use any other work (either published or online)
IMAGE PROCESSING SYSTEMS

Image processing systems refer to computer-

based systems or software applications
designed to manipulate, analyze, enhance,
and interpret digital images. Image acquisition
These
The pixel systems
bit depthare used
is the in a of
number wide
bits range of been made available in the
that have Image enhancement
applicationsdigitalacross various industries,
system to represent each pixel in the image.
including Here healthcare, entertainment,
we have an example of using only four bits. Feature extraction
manufacturing, security, and more.
Image segmentation
Here are key components and functionalities
This is smaller than would be used in any actual medical image because with
fourofbits,
image processing
a pixel would besystems: Object detection
limited to having only 16 different values (brightness
levels or shades of gray). Scene understanding

Add a Footer 5
 DIGITAL IMAGES • An image, such as a photograph or x-ray, which is stored
• An Image is defined as a two dimensional function f(x,y) exactly as recorded by the camera or x-ray film, as a
where x and y are spatial coordinates and amplitude of ‘f’ “picture”. In this form, the image can only be retrieved
at any coordinates (x,y) is the intensity value or gray level as the picture of the object, text page, or graphic; it
of image at that point. cannot be retrieved by electronic searching of the image
itself, nor, if text, can its contents be handled
• Image is a digital image when the spatial coordinates x,y electronically
and intensity value of f’ have finite and discrete quantities. • Analog images are the type of images that we,
• A digital image is an array of real or complex numbers as humans, look at.  They include such things as
represented by a finite number so fo bits. photographs, paintings, TV images, and all of our
medical images recorded on film or displayed on various
display devices, like computer monitors.  What we see in
an analog image is various levels of brightness (or film
density) and colors.  It is generally continuous and not
broken into many small individual pieces.
• To retrieve the analog image electronically, there must
be a separate and parallel system in digital form, which
indexes the analog image in some fashion.

Source: http://www.sprawls.org/resources/DICHAR/module.htm
IMAGE PROCESSING SYSTEMS

Image Acquisition: Image processing systems start with the

acquisition of digital images from various sources, such as cameras,
scanners, satellites, or medical imaging devices. The quality and type
of input data greatly influence subsequent processing steps.
The pixel bit depth is the number of bits that have been made available in the
digital system to represent each pixel in the image.
Here we
• A digital image is ahave an example of
representation of ausing
two-only four bits.
dimensional image as a finite set of digital values,
This is smaller
called thanelements
picture would beor used in any actual medical image because with
pixels
four bits, a pixel would be limited to having only 16 different values (brightness
• Pixel values typically represent gray levels, colors,
levels or shades of gray).
heights, opacities etc
• Remember digitization implies that a digital image
is an approximation of a real scene

Add a Footer 7
 IMAGE SAMPLING AND QUANTIZATION
• Digitizing the coordinate value is called sampling and
digitizing the amplitude values is called quantization.
• After sampling, the values of the samples span a
continuous range of intensity values must be converted int
to discrete quantities. This is called quantization.
• A quantizer maps a continuous variable t in a discrete
variable r. This mapping is a staircase function and the
quantizer follows quantization rules.
 The type of quantization in which the quantization
levels are uniformly spaced is termed as a Uniform
Quantization.
 The type of quantization in which the quantization https://www.sci.utah.edu/~arpaiva/classes/UT_ece6962/
image_representation_and_discretization.pdf
levels are unequal and mostly the relation between
them is logarithmic, is termed as a Non-uniform
Quantization.

Source: http://www.sprawls.org/resources/DICHAR/module.htm
 IMAGE QUANTIZATION

The difference between an input value and its

quantized value is called a Quantization
Error.
https://www.sci.utah.edu/~arpaiva/classes/UT_ece6962/image_representation_and_discretization.pdf
 IMAGE RECONSTRUCTION
Image reconstruction is the process of creating a visual representation or image from a set of data
samples or measurements.
This process is commonly used in various fields, including medical imaging (e.g., CT scans, MRI), remote
sensing (e.g., satellite imagery).
Image reconstruction aims to recover an image that accurately represents the underlying object or scene
from which the data samples were acquired.

LiDAR point cloud data view

showing individual 3D feature
level information. Individual
points are color coded by
Mean Sea Level (MSL) height.

Source: 10.1117/12.2049856
 DIGITAL IMAGE REPRESENTATION
• An image is an array, or a matrix, of square pixels (picture
elements) arranged in columns and rows
• A digital image is represented in the imaging and computer
system by numbers in the form of binary digits, called bits.
The• pixel
Herebit
wedepth
see theisgeneral
the number ofofbits
structure that have
a digital image.been made available in the
• First, it isdigital
dividedsystem to represent
into a matrix of pixels.each pixel in the image.
Here we have an example of using only four bits.
• Then, each pixel is represented a series of bits.
This is smaller than would be used in any actual medical image because with
four bits, a pixel would be limited to having only 16 different values (brightness
The pixel bit depth is thelevels or shades
number of bits of gray).
that have been made
available in the digital system to represent each pixel in the
image.
Here we have an example of using only four bits.
This is smaller than would be used in any actual medical image
because with four bits, a pixel would be limited to having only
16 different values (brightness levels or shades of gray).
 DIGITAL IMAGE – BIT SYSTEMS
• Number of Shades or brightness levels at a given wavelength

• Smallest change in
intensity level that can
be detected by the
sensing system
When the pixel bit depth is increased to eight bits, a pixel can then have 256
different values (brightness levels, shades of gray, etc)
The pixel bit depth is the number of bits that have been made available in the
digital system to represent each pixel in the image.
Here we have an example of using only four bits.

This is smaller than would be used in any actual medical image because with
four bits, a pixel would be limited to having only 16 different values (brightness
levels or shades of gray).

Add a Footer 12
 Spatial Resolution
PIXEL SIZE AND DIGITAL IMAGE DETAIL
• When an image is in digital form, it is actually blurred by the size of the pixel. This is because all
anatomical detail within an individual pixel is "blurred together" and represented by one number.
• The physical size of a pixel, relative to the anatomical objects, is the amount of blurring added to the
imaging process by the digitizing of the image.
• Here we see that an image with small pixels (less blurring) displays much more detail than an image
made
The pixel bit up of larger
depth is thepixels.
number of bits that have been made available in the
digital system to represent each pixel in the image.
Here we have an example of using only four bits.

This is smaller than would be used in any actual medical image because with
four bits, a pixel would be limited to having only 16 different values (brightness
levels or shades of gray).

30m resolution 15m resolution 1m resolution

Add a Footer 13
 DIGITAL IMAGE DETAIL
• Unlike grayscale images, which represent pixel values as shades of gray, color images use a
combination of colors to represent the visual information.
• Color images are typically represented using one of the following color models:

RGB (Red, Green, Blue): each pixel is represented by three values:

the amount of red, green, and blue light required to create the
The pixel
color ofbit depth
that pixel.isBy
thevarying
number theofintensity
bits thatof
have been
these made
three available in the
primary
digital
colors, a wide rangesystem to represent
of colors each pixelIt's
can be generated. in the
alsoimage.
known as
Here we have an example
a "true color" or "full-color" image. of using only four bits.

CMYK
This (Cyan, than
is smaller Magenta,
wouldYellow,
be usedKey/Black): colors
in any actual are created
medical by
image because with
subtracting
four different
bits, a pixel percentages
would be limited to of cyan,only
having magenta, yellow,values
16 different and (brightness
black ink from a white background. CMYKofisgray).
levels or shades used for color
reproduction in printed materials.
HSV (Hue, Saturation, Value): The HSV color model represents YUV/YCbCr: YUV and YCbCr color models
colors in terms of their hue (the type of color, like red or blue), separate the luminance (brightness) component
saturation (the intensity or purity of the color), and value (the (Y) from the chrominance (color) components (U
brightness or darkness of the color). It's often used in image and V or Cb and Cr). These models are
processing for tasks like color manipulation and segmentation. commonly used in video compression and
transmission, as they allow for efficient
Add a Footer representation of color information. 14
 IMAGE PROCESSING SYSTEMS
Image Enhancement: Enhancement techniques
aim to improve the visual quality of images.
They include operations like contrast stretching,
brightness adjustment, histogram equalization,
and sharpening. Image enhancement is used to
The pixel
make bit depth
features moreis the number
visible or to of bits that
improve thehave been made available in the
digital system
overall appearance to represent each pixel in the image.
of images.
Here we have an example of using only four bits.

This is smaller than would be used in any actual medical image because with
four bits, a pixel would be limited to having only 16 different values (brightness
levels or shades of gray).

Image Restoration: Image restoration is the process of

removing or reducing degradations, such as blurriness or noise,
that occur during image acquisition. Deconvolution and various
filtering techniques are commonly used for image restoration.
Source: 10.5194/hess-21-1017-2017

Add a Footer 15
 IMAGE PROCESSING SYSTEMS
Feature Extraction: Feature extraction
involves identifying and extracting meaningful
information or features from images. These
features could include edges, corners, textures,
shapes, colors, or any other relevant
The pixel bit depth
information. Featureis the numberisofcrucial
extraction bits that
forhave been made available in the
digitalrecognition,
object detection, system to represent each pixel in the image.
and analysis.
Feature TypesHere we have an example of using only four bits. https://www.nv5geospatialsoftware.com/docs/FXRuleBasedTutorial.html
Spectral Features: These features are derived from the spectral characteristics of the data. Examples include spectral
This is smaller than would be used in any actual medical image because with
indices (e.g., NDVI, NDWI), band ratios, and principal component analysis (PCA) components.
four bits, a pixel would be limited to having only 16 different values (brightness
Textural Features: These features capture patterns in pixel intensity or spectral values and are useful for
levels or shades of gray).
characterizing texture in the imagery.
Spatial Features: Spatial features describe the arrangement and relationships between objects or pixels. Examples
include object size, shape, and proximity to other features.
Statistical Features: These features encompass statistical measures of data distribution, such as mean, standard
deviation, and skewness.
Geometric Features: Geometric features describe the geometry or shape of objects in the imagery, often used in
object recognition tasks.

Add a Footer 16
 IMAGE PROCESSING SYSTEMS
Segmentation is the process of dividing image into
meaningful and homogeneous regions or segments based
on certain criteria, such as color, texture, or spectral
characteristics. The goal of segmentation is to partition an
image into distinct areas that represent objects, land cover
types, or other significant features on the Earth's surface.
The pixel bit depth
Segmentation is theinnumber
is is used variousofapplications,
bits that have been made available in the
including
digital system
land cover classification, to represent
change each
detection, andpixel in the image.
object
recognition. Here we have an example of using only four bits. https://www.hindawi.com/journals/mpe/2021/2472726/
Object
This is detection in remote
smaller than sensing
would be refers
used in any to the process
actual medical ofimage because with
identifying and locating specific objects or features of
four bits, a pixel would be limited to having only 16 different values (brightness
interest within satellite or aerial
levelsimagery.
or shadesThis capability is
of gray).
essential for various applications, including urban planning,
agriculture, environmental monitoring, disaster assessment,
and more. Number of object detection techniques are
available. The choice of method and approach depends on
factors like the complexity of the objects, the quality of the
imagery, and the availability of training data. Deep learning
techniques such as template matching, edge detection,
machine learning etc.
Add a Footer https://www.sciencedirect.com/science/article/abs/pii/S0924271619302825 17
 IMAGE PROCESSING SYSTEMS
Visualization and reporting involves the
presentation and communication of processed
remote sensing data, analysis results, and
findings in a comprehensible and informative
manner.
The pixel bitstakeholders
It enables depth is the tonumber of bits that have been made available in the
make informed
digital systemfindings
decisions, communicate to represent
to the each pixel in the image.
public,
Here we have
and support scientific an example
research and of using only four bits.
environmental management. They play a vital
This
roleisinsmaller than would
transforming be used
complex data in any actual medical image because with
into
four bits, a pixel
actionable would be limited to having only 16 different values (brightness
insights.
levels or shades of gray).

Forest fire pattern and vulnerability

mapping using deep learning in
Nepal
https://fireecology.springeropen.com/articles/10.1186/s42408-022-00162-3

Add a Footer 18
THANK YOU

19