1. Explain the components of image processing.

(UNIT 1)
2. Explain the various enhancement techniques performed in spatial domain (UNIT-2)
3. Explain region splitting and merging technique foe image segmentation with suitable
examples (UNIT 3)
4. Encode the sentence ‘ I LOVE IMAGE PROCESSING ‘ using arithmetic coding procedure
5. Explain in detail about the object recognition technique based on matching (UNIT 3)
6. Explain the various boundary descriptors details with a neat diagram (UNIT 3)

ANS:

Q1. Components of Image Processing System (UNIT 1)

It is is the combination of the different elements involved in the digital image
processing. Digital image processing is the processing of an image by means of a
digital computer. Digital image processing uses different computer algorithms to
perform image processing on the digital images.
It consists of following components:-
1. Image Sensors:
Image sensors senses the intensity, amplitude, co-ordinates and other features of the
images and passes the result to the Image Processing Hardware:
Image processing hardware is the dedicated hardware that is used to process the
instructions obtained from the image sensors. It passes the result to general purpose

computer.
2. Computer:
Computer used in the image processing system is the general purpose computer that
is used by us in our daily life.
3. Image Processing Software:
Image processing software is the software that includes all the mechanisms and
algorithms that are used in image processing system.
4. Mass Storage:
Mass storage stores the pixels of the images during the processing.
5. Hard Copy Device:
Once the image is processed then it is stored in the hard copy device. It can be a pen
drive or any external ROM device.
6. Image Display:
It includes the monitor or display screen that displays the processed images.
7. Network:
Network is the connection of all the above elements of the image processing system.
Q2. Explain the various enhancement techniques performed in spatial
domain
In the spatial domain, image enhancement techniques aim to improve the quality or the
interpretability of an image by altering its pixel values directly. The primary goal is to make
specific details of the image more visible or aesthetically pleasing. Here are the various
enhancement techniques in the spatial domain:
1. Contrast Enhancement
● Linear Contrast Stretching (Normalization): This technique enhances the
contrast by stretching the pixel intensity values over the entire available range.

● Histogram Equalization: This technique enhances the image contrast by

redistributing the intensity values so that they are spread more evenly over the
available range. The goal is to make the histogram of the image more uniform.
o Involves computing the cumulative distribution function (CDF) and mapping the
original pixel values to new values based on the CDF.

2. Brightness Enhancement

● Addition or Subtraction of Constant Value: This technique enhances the image by adding or
subtracting a constant value from all the pixels, increasing or decreasing the brightness of
the entire image.

Gamma Correction: This technique involves adjusting the brightness of an image using a power-law
relationship. It can be used to either brighten or darken the image depending on the value of the
gamma parameter.

3. Smoothing (Blurring)
Smoothing techniques are applied to reduce noise and fine details in an image, typically to
make the image more uniform or to reduce high-frequency noise.
 ● Average Filtering: This technique involves averaging the pixel values within a
neighborhood around each pixel, effectively smoothing the image.

4. Edge Enhancement
Edge enhancement techniques emphasize boundaries and transitions between regions of
different intensities, highlighting the features in an image.

● High-Pass Filtering: High-pass filters are used to emphasize high-frequency components in

the image, which correspond to edges and fine details. The image is convolved with a kernel
that highlights high-frequency regions.

Unsharp Masking: This technique enhances edges by subtracting a

blurred version of the image from the original image. The resulting image is a sharper
version of the original image.

5. Sharpening
Sharpening techniques enhance the fine details and edges of the image by emphasizing high-
frequency components, making the image look more distinct.
● Laplacian Sharpening: This is done by subtracting the Laplacian-filtered image from
the original image, amplifying the high-frequency components (edges).
● High-Boost Filtering: High-boost filtering is a modification of unsharp masking. It
involves adding a scaled version of the high-pass filtered image back to the original
image.

Q3. Explain region splitting and merging technique foe image

segmentation with suitable examples
Region Splitting and Merging is a popular technique used for image segmentation,
which divides an image into multiple segments (regions) based on certain homogeneity
criteria, such as intensity, texture, or color. This method operates in two primary phases:
region splitting and region merging.
1. Region Splitting
In the region splitting phase, the entire image is initially considered as a single region. The
goal is to break this region into smaller sub-regions based on some predefined criteria (e.g.,
intensity, color, or texture homogeneity). If a region does not meet the homogeneity criterion,
it is split into smaller sub-regions. This process continues recursively until all regions meet
the homogeneity criteria.
Process:

● The entire image is treated as a large initial region.

● The image is split recursively into smaller sub-regions based on some similarity or
homogeneity criterion.
● The splitting stops when all the sub-regions are homogeneous or sufficiently small.

Example of Region Splitting:

Consider an image with different objects (e.g., a landscape containing sky, ground, and trees).
Initially, the entire image is treated as one region. If the pixel intensities of the sky and the
ground are significantly different, the image is split into two regions: one for the sky and one
for the ground. This splitting may continue further to separate the trees from the ground,
depending on the homogeneity criterion.

2. Region Merging
After the image is split into smaller regions in the splitting phase, the merging phase aims to
combine adjacent regions that are similar or homogeneous according to the predefined
criterion. Merging is done to avoid overly fragmented or small regions that may not be useful
for further analysis.
Process:

● Initially, regions that were split in the previous phase are considered.
● Pairs of adjacent regions are examined to see if they meet the homogeneity criteria.
● If adjacent regions are similar enough, they are merged back together.
● The process is repeated iteratively until no further merging is possible.
 ●

Example of Region Merging:

After splitting the image into multiple small regions, such as individual tree branches, you
may observe that some small segments in the branches have similar pixel intensities and thus
could be merged into a single region representing a larger branch. The merging process
would combine these similar regions back together to form more meaningful segments.
Key Steps in Region Splitting and Merging

1. Initial Region: Start with the entire image as a single region.

2. Splitting: Recursively split the image into smaller regions based on homogeneity criteria.
3. Merging: After splitting, merge adjacent homogeneous regions to form larger meaningful
segments.

Q4. Encode the sentence ‘ I LOVE IMAGE PROCESSING ‘ using arithmetic coding procedure

Here’s the step-by-step procedure to encode the sentence "I LOVE IMAGE
PROCESSING" using Arithmetic Coding.
Step 1: Analyze the Input Data
We begin by analyzing the frequency of each character in the sentence. The sentence is "I
LOVE IMAGE PROCESSING". First, remove the spaces for simplicity in this explanation
(although they will be included in the final coding).
The sequence of characters is:
arduino
Copy code
"ILOVEIMAGEPROCESSING"

We need to calculate the frequency of each character.

Charact Frequenc
er y
I 2
L 1
O 2
V 1
E 2
M 1
A 1
G 1
P 1
R 2
C 1
S 2
 Charact Frequenc
er y
N 1

Step 2: Calculate the Probabilities

Next, calculate the probability of each character in the string by dividing the frequency of
each character by the total number of characters.
The total number of characters is 18 (counting spaces as characters):
java
Copy code
Total characters = 18
Now, calculate the probabilities of each character.
Charact Frequen
Probability
er cy

2/18 =
I 2
0.111

1/18 =
L 1
0.056

2/18 =
O 2
0.111

1/18 =
V 1
0.056

2/18 =
E 2
0.111

1/18 =
M 1
0.056

1/18 =
A 1
0.056

1/18 =
G 1
0.056

1/18 =
P 1
0.056

2/18 =
R 2
0.111

1/18 =
C 1
0.056

2/18 =
S 2
0.111

N 1 1/18 =
 Charact Frequen
Probability
er cy

0.056

Step 3: Assign Cumulative Ranges to Each Symbol

In this step, we need to assign cumulative probability intervals to each symbol based on the
probabilities calculated. We first sort the characters based on their probability in ascending
order.
Sorted Characters by Probability:
Charact Probabili Cumulative
er ty Range

L 0.056 [0.000, 0.056)

V 0.056 [0.056, 0.112)

M 0.056 [0.112, 0.168)

A 0.056 [0.168, 0.224)

G 0.056 [0.224, 0.280)

P 0.056 [0.280, 0.336)

C 0.056 [0.336, 0.392)

N 0.056 [0.392, 0.448)

I 0.111 [0.448, 0.559)

O 0.111 [0.559, 0.670)

E 0.111 [0.670, 0.781)

R 0.111 [0.781, 0.892)

S 0.111 [0.892, 1.000)

Step 4: Start Encoding

The encoding procedure involves maintaining a range [low,high][low, high][low,high] which
initially starts as [0.0,1.0)[0.0, 1.0)[0.0,1.0). For each character in the message, the range is
refined based on the cumulative probability range assigned to that character.
For the message "I LOVE IMAGE PROCESSING", we go through each character and update
the range. The general procedure is:

1. For each character in the message:

o Find the cumulative range for the character.
o Update the low and high values of the range to correspond to this cumulative range.

Example for the first few characters:

● Character 1: I
 o Range for "I" = [0.448,0.559)[0.448, 0.559)[0.448,0.559)
o New range: [low,high]=[0.448,0.559)[low, high] = [0.448, 0.559)
[low,high]=[0.448,0.559)
● Character 2: L
o Range for "L" = [0.000,0.056)[0.000, 0.056)[0.000,0.056)
o New range: Update to
[low,high]=[0.448+(0.559−0.448)×0.000,0.448+(0.559−0.448)×0.056)[low, high] =
[0.448 + (0.559 - 0.448) \times 0.000, 0.448 + (0.559 - 0.448) \times 0.056)
[low,high]=[0.448+(0.559−0.448)×0.000,0.448+(0.559−0.448)×0.056)
o Which results in the new range: [0.448,0.453)[0.448, 0.453)[0.448,0.453)

The process continues for each character until the entire string is encoded.
Q5. Explain in detail about the object recognition technique based on
matching
Object Recognition Based on Matching is a technique used to identify objects in an image
by comparing features or patterns of the object in the image with a pre-defined set of
templates or models stored in a database. This method of object recognition relies heavily on
the concept of matching specific parts of the image with a model or template, and it is widely
used in computer vision and image processing applications such as robotics, medical
imaging, autonomous vehicles, and augmented reality.
Here’s a detailed explanation of the object recognition technique based on matching:
1. Overview of Object Recognition
Object recognition aims to locate and identify objects within an image. In matching-based
object recognition, we compare the image features of the object with predefined models
(which may be templates or feature descriptors). The goal is to determine which model
matches the image the best.
2. Types of Matching-Based Object Recognition
There are two main categories of matching-based object recognition:
● Template Matching: This is a basic form of object recognition where predefined
templates (or models) of the object are matched against portions of the image.
● Feature-Based Matching: This method involves extracting features from the image
(e.g., key points, edges, textures) and comparing these features to those of a template
or a known model of the object. Feature-based matching often handles more complex
images and variations like rotation, scaling, and illumination changes.
3. Template Matching
Template matching is a straightforward matching-based recognition technique where we
compare the entire image with a template of the object.
Steps in Template Matching:
1. Template Creation: First, you create a template image that contains the object you
want to recognize. This is a simple image, often in grayscale, of the object of interest.
2. Sliding Window: The template is moved (or "slid") across the entire image in a pixel-
by-pixel manner. At each position, the similarity between the template and the current
region of the image is calculated.
 3. Similarity Measurement: A similarity or dissimilarity measure is used to compare
the template and the image region. Common similarity measures include:
o Sum of Squared Differences (SSD): Measures the squared difference between
corresponding pixel values of the template and the image region.
o Normalized Cross-Correlation (NCC): Measures how well the template and the
image region match, considering both intensity and correlation.
4. Best Match Identification: After sliding the template over the image, the position
with the highest similarity score is considered the best match. This position is where
the object is most likely located.
Advantages of Template Matching:

● Simple to Implement: It is easy to implement and computationally simple for small images
or objects.
● Effective for Fixed Objects: It works well when the objects are in a fixed orientation and
have minimal changes in scale or lighting.

Disadvantages of Template Matching:

● Sensitivity to Changes: It is sensitive to changes in scale, rotation, or viewpoint. If the object

in the image is rotated, scaled, or has different lighting conditions, the method might not
work well.
● Computationally Expensive: For large images or multiple templates, sliding the template
over every part of the image can be computationally expensive.

4. Feature-Based Matching
Feature-based matching is more sophisticated than template matching and is particularly
useful for recognizing objects in more complex and varied environments, as it can handle
transformations such as rotation, scaling, and partial occlusion.
Key Steps in Feature-Based Matching:
1. Feature Extraction: The first step is to extract distinctive features from both the
object (template) and the image. These features may include:
o Keypoints: Points in the image that are distinct and easily detectable (e.g., corners,
edges).
o Descriptors: Descriptions of the features that allow comparison between different
keypoints (e.g., SIFT, SURF, ORB).
2. Feature Matching: Once the features are extracted, we compare the features of the
object (template) with those extracted from the image. This involves matching the
descriptors of the keypoints in the template with those of the keypoints in the image.
Several techniques are used to match these features:
o Brute Force Matching: Involves comparing each feature in the template with every
feature in the image.
o Approximate Nearest Neighbor (ANN): A more efficient approach to find the closest
matching features in a larger image, often using data structures like KD-Trees.
3. Geometric Transformation: Once the features are matched, the next step is to
estimate the geometric transformation that maps the object’s features to those in the
image. This step involves calculating a transformation matrix that accounts for:
o Scaling
 o Rotation
o Translation
o Affine Transformations (if necessary)

This is achieved by using algorithms like RANSAC (Random Sample Consensus),

which helps find the best transformation by ignoring outliers in the matched features.
4. Object Localization: After the transformation is computed, the object can be
localized in the image. The coordinates of the matched features are used to draw a
bounding box around the object or to mark the location of the object in the image.
Popular Feature-Based Matching Algorithms:

● SIFT (Scale-Invariant Feature Transform): Detects and describes local features in images
that are invariant to scaling, rotation, and affine transformations.
● SURF (Speeded-Up Robust Features): Similar to SIFT but optimized for faster computation.
● ORB (Oriented FAST and Rotated BRIEF): A fast and efficient alternative to SIFT and SURF,
particularly useful for real-time applications.
● FAST (Features from Accelerated Segment Test): A corner detection algorithm used for
keypoint extraction.

Advantages of Feature-Based Matching:

● Invariance to Transformations: It can handle transformations like rotation, scaling, and
changes in viewpoint.
● Robustness: It works well in scenarios with partial occlusion or cluttered backgrounds.
● Efficiency: Many feature-based methods are computationally efficient, even in large images.
Disadvantages of Feature-Based Matching:
● Complexity: The feature extraction and matching process can be computationally complex
compared to template matching.
● Sensitivity to Noise: Some methods may be sensitive to noise in the image, which can result
in incorrect feature matching.

Boundary descriptors are essential tools used in image processing and computer vision to
describe the shape, outline, or contour of objects within an image. These descriptors are
useful for tasks like object recognition, shape analysis, and pattern matching. The idea behind
boundary descriptors is to extract relevant features from an object's boundary that can be used
to represent the object efficiently.
Boundary descriptors are typically applied to binary images where the object of interest is
separated from the background. The boundary of an object is the set of pixels that forms the
outline of the object, typically denoted by a curve or a set of connected points.
Various Boundary Descriptors
Here are some of the commonly used boundary descriptors:
1. Chain Codes
Chain code is one of the simplest and earliest methods for representing object boundaries. It
encodes the boundary of an object as a sequence of directions or steps, which describe the
path along the boundary starting from an initial point.
Process:
● The boundary is traced using 8-connectivity (or 4-connectivity).
● At each step, a direction is assigned based on the movement along the boundary.
● A direction is typically encoded as one of 8 directions (0 to 7), where each number
represents a specific angle or direction.
Example:
 ● Chain Code of a boundary: For example, if the object’s boundary moves "up-right", "right",
"down-right", etc., the chain code could look something like [2, 3, 4].

Fourier Descriptors
Fourier descriptors are used to represent the shape of an object boundary by transforming the
boundary into the frequency domain. The idea is to represent the boundary in terms of its
Fourier series, which captures the object shape in terms of sinusoidal components (sine and
cosine waves).
Process:
● First, the boundary is converted into a set of complex numbers by applying a Fourier
transform.
● The Fourier coefficients represent the boundary shape in terms of frequency.
● By truncating higher frequencies, the Fourier descriptor provides a compact representation
of the boundary.
Formula:
● The boundary points are mapped to a complex plane: Z(k)=x(k)+iy(k)Z(k) = x(k) + i
y(k)Z(k)=x(k)+iy(k), where x(k)x(k)x(k) and y(k)y(k)y(k) are the coordinates of boundary
points.
● Apply the Discrete Fourier Transform (DFT) to the boundary points.
3. Polygonal Approximation
Polygonal approximation is a method where the boundary of an object is approximated by a
polygon with fewer vertices, making the shape representation simpler and more compact.
Process:
● The boundary is traced, and the direction of the boundary is calculated.
● The boundary is then approximated by connecting points with straight lines, replacing curves
with straight edges (using algorithms like the Douglas-Peucker algorithm).
Example:
● A circular boundary might be approximated as a polygon with a few vertices (e.g., a hexagon
or octagon) that captures the overall shape.

. Moment Invariants
Moment invariants are used to describe the shape of an object’s boundary through
mathematical moments. Moments are integrals of a function that can represent different
properties of the boundary, such as area, center of mass, and orientation.

Types of Moments:
● Geometric Moments: Moments that are calculated based on the geometric properties of
the object.
● Hu Moments: A set of seven moment invariants that are invariant to translation, scaling, and
rotation.

5. Boundary Fourier Transform

This technique involves applying the Fourier transform directly to the boundary, rather than
using the chain code. The result is a set of Fourier coefficients that describe the shape of the
object boundary. It is very similar to Fourier descriptors but applies specifically to the
boundary representation.
Process:
● Boundary points are represented as a sequence of 2D coordinates.
● Fourier transform is applied to this sequence, converting the boundary into a series of
frequency components.
6. Curvature Scale Space (CSS) Representation
Curvature Scale Space (CSS) is a method used to capture the curvature of the boundary at
various scales. The boundary of an object is represented by its curvature at different scales,
helping to recognize objects despite noise, changes in scale, and viewpoint.
Process:
● The boundary is smoothed at different scales, and the curvature at each scale is calculated.
● The resulting representation provides a multiscale description of the boundary that is less
sensitive to small fluctuations or noise.
Advantages:
● Robust to noise and occlusions.
● Invariant to transformations like scaling and rotation.