IMAGE PROCESSING

Assignment 1

Name: SHWETA TIWARI

Roll Number: 2204280100168

Branch: CSE- (3rd Year)

Ques 1: Explain sampling and quantization. Explain the
effects of reducing sampling and quantization.
Answer:
Sampling:

1. Sampling refers to measuring the intensity values of an image at regular intervals (dis-
crete points).

2. It involves selecting the number of samples in each dimension (horizontal and vertical).

3. A higher sampling rate leads to better spatial resolution.

4. Low sampling rates can cause aliasing, resulting in a loss of image details.

5. Example: Reducing the number of pixels sampled from an image decreases its resolution.

Quantization:

1. Quantization is the process of mapping continuous pixel values into discrete levels.

2. It reduces the number of distinct gray levels available for an image.

3. A higher number of quantization levels means better quality and fewer visible artifacts.

4. Low quantization levels can cause contouring effects, where smooth gradients appear
as bands.

5. Example: Reducing an 8-bit image (256 levels) to a 4-bit image (16 levels).

Effects of Reducing Sampling and Quantization:

1. Reduced Sampling Rate leads to loss of spatial detail and aliasing.

2. Reduced Quantization Levels result in visible quantization noise and contouring.

3. Loss of Image Details: Reduces the amount of information, leading to loss of finer
details.

4. Increased Distortion due to artifacts that distort the image.

5. Smaller File Size: Reduces storage requirements.

1
Ques 2: What do you mean by image processing? Explain
the steps in image processing with the help of a block
diagram.
Answer:
Image Processing:

1. Image processing involves operations that transform an image into a form suitable for
analysis.

2. It includes tasks like enhancing, restoring, and analyzing images.

3. Common applications include medical imaging, computer vision, and remote sensing.

Steps in Image Processing:

1. Image Acquisition: Capturing an image using a sensor.

2. Preprocessing: Reducing noise and enhancing image quality.

3. Segmentation: Dividing the image into meaningful regions.

4. Feature Extraction: Identifying important features like edges.

5. Classification: Assigning labels to features.

6. Post-Processing: Refining the results.

7. Output: Displaying or storing the processed image.

2
Ques 3: Explain the steps involved in sampling and digi-
tization of images. How many minutes are required for a
512 x 512 image with 256 grey levels at 300 baud rate for
transmission?
Answer:
Sampling and Digitization Steps:

1. Sampling: Converting a continuous image into a discrete grid of pixels.

2. Quantization: Mapping pixel values to a limited number of gray levels.

3. Digitization: Representing image data in binary form for storage and processing.

Transmission Calculation:

1. Image size = 512 x 512 pixels.

2. Each pixel requires 8 bits (256 grey levels).

3. Total bits = 512 x 512 x 8 = 2,097,152 bits.

4. Transmission rate = 300 bits/second.

5. Time = Total bits / Transmission rate = 2,097,152 / 300.

6. Time in minutes = (2,097,152 / 300) / 60 = approximately 116.5 minutes.

3
Ques 4: Compute f + h, f ∗ h, f ◦ h, f · h for the sequences
f = [5, 7, 11, 8, 2, 6, 8, 9, 7, 4, 3], h = [1, 2, 1].
Answer:
Given Sequences:
f = [5, 7, 11, 8, 2, 6, 8, 9, 7, 4, 3]
h = [1, 2, 1]
Computations:

1. Element-wise Addition f + h:
• The lengths of f and h are different, so we will add h to the first few elements of
f and keep the remaining elements of f unchanged.
• Resulting sequence:

f + h = [5 + 1, 7 + 2, 11 + 1, 8, 2, 6, 8, 9, 7, 4, 3] = [6, 9, 12, 8, 2, 6, 8, 9, 7, 4, 3]

2. Convolution f ∗ h:
• The convolution of two sequences is calculated as:
M
X
(f ∗ h)[n] = f [m]h[n − m]
m=0

where M is the length of h.

• The result will have a length of N + M − 1 = 11 + 3 − 1 = 13.
• Computing convolution: (f ∗h)[0] = 5 · 1 = 5
(f ∗ h)[1] = 5 · 2 + 7 · 1 = 10 + 7 = 17
(f ∗ h)[2] = 5 · 1 + 7 · 2 + 11 · 1 = 5 + 14 + 11 = 30
(f ∗ h)[3] = 7 · 1 + 11 · 2 + 8 · 1 = 7 + 22 + 8 = 37
(f ∗ h)[4] = 11 · 1 + 8 · 2 + 2 · 1 = 11 + 16 + 2 = 29
(f ∗ h)[5] = 8 · 1 + 2 · 2 + 6 · 1 = 8 + 4 + 6 = 18
(f ∗ h)[6] = 2 · 1 + 6 · 2 + 8 · 1 = 2 + 12 + 8 = 22
(f ∗ h)[7] = 6 · 1 + 8 · 2 + 9 · 1 = 6 + 16 + 9 = 31
(f ∗ h)[8] = 8 · 1 + 9 · 2 + 7 · 1 = 8 + 18 + 7 = 33
(f ∗ h)[9] = 9 · 1 + 7 · 2 + 4 · 1 = 9 + 14 + 4 = 27
(f ∗ h)[10] = 7 · 1 + 4 · 2 + 3 · 1 = 7 + 8 + 3 = 18
(f ∗ h)[11] = 4 · 1 + 3 · 2 = 4 + 6 = 10
(f ∗ h)[12] = 3 · 1 = 3

• Final convolution result:

f ∗ h = [5, 17, 30, 37, 29, 18, 22, 31, 33, 27, 18, 10, 3]

3. Circular Convolution f ◦ h:
• Circular convolution is similar to regular convolution, but the indices wrap around
at the ends.

4
 • The result will also be of length N (11).
• Computing circular convolution: (f ◦h)[0] = 5 · 1 + 3 · 2 + 7 · 1 = 5 + 6 + 7 = 18
(f ◦ h)[1] = 5 · 2 + 7 · 1 + 11 · 1 = 10 + 7 + 11 = 28
(f ◦ h)[2] = 7 · 2 + 11 · 1 + 8 · 1 = 14 + 11 + 8 = 33
(f ◦ h)[3] = 11 · 2 + 8 · 1 + 2 · 1 = 22 + 8 + 2 = 32
(f ◦ h)[4] = 8 · 2 + 2 · 1 + 6 · 1 = 16 + 2 + 6 = 24
(f ◦ h)[5] = 2 · 2 + 6 · 1 + 8 · 1 = 4 + 6 + 8 = 18
(f ◦ h)[6] = 6 · 2 + 8 · 1 + 9 · 1 = 12 + 8 + 9 = 29
(f ◦ h)[7] = 8 · 2 + 9 · 1 + 7 · 1 = 16 + 9 + 7 = 32
(f ◦ h)[8] = 9 · 2 + 7 · 1 + 4 · 1 = 18 + 7 + 4 = 29
(f ◦ h)[9] = 7 · 2 + 4 · 1 + 3 · 1 = 14 + 4 + 3 = 21
(f ◦ h)[10] = 4 · 2 + 3 · 1 + 5 · 1 = 8 + 3 + 5 = 16

• Final circular convolution result:

f ◦ h = [18, 28, 33, 32, 24, 18, 29, 32, 29, 21, 16]

4. Dot Product f · h:

• The dot product is computed as:

M
X −1
f ·h= f [n] · h[n]
n=0

where M is the length of h.

• The lengths of f and h are different, so we only compute for the overlapping
values: f ·h = f [0]h[0] + f [1]h[1] + f [2]h[2]
= 5 · 1 + 7 · 2 + 11 · 1
= 5 + 14 + 11 = 30
• Final dot product result:
f · h = 30

Final Results:
f + h = [6, 9, 12, 8, 2, 6, 8, 9, 7, 4, 3]
f ∗ h = [5, 17, 30, 37, 29, 18, 22, 31, 33, 27, 18, 10, 3]
f ◦ h = [18, 28, 33, 32, 24, 18, 29, 32, 29, 21, 16]
f · h = 30

5
Ques 5: Describe in detail the elements of digital image
processing system and describe Sampling and Quantiza-
tion.
Answer:
Elements of Digital Image Processing:

1. Image Acquisition: Capturing images using sensors.

2. Image Storage: Storing images in digital formats.

3. Image Processing Hardware: Specialized hardware for image manipulation.

4. Image Processing Software: Software tools for image analysis.

5. Display Devices: Visualizing processed images.

6. Image Printing: Producing hard copies.

7. Feedback Mechanism: Adjusting processes based on output.

6
Ques 6: Explain the 4, 8, and m connectivity of pixels.
Explain region, edge in context with connectivity of pixels.
Answer:
Connectivity of Pixels:

1. 4-Connectivity: A pixel is connected to its horizontal and vertical neighbors. A pixel

at (x, y) is connected to (x + 1, y), (x − 1, y), (x, y + 1), and (x, y − 1).

2. 8-Connectivity: A pixel is connected to its horizontal, vertical, and diagonal neighbors.

A pixel at (x, y) is connected to (x+1, y), (x−1, y), (x, y+1), (x, y−1), and diagonally
adjacent pixels.

3. m-Connectivity: A mix of 4 and 8 connectivity to avoid ambiguity in connected regions.

It prioritizes 4-connectivity unless both diagonally connected pixels are part of the same
region.

Region and Edge:

1. Region: A set of connected pixels with similar properties, forming an area in the image.

2. Edge: A boundary between two regions with different properties, such as intensity or
color.

3. 4-Connectivity Edge: Boundaries formed using 4-connected pixels.

4. 8-Connectivity Edge: Boundaries formed using 8-connected pixels, including diagonal

connections.

7
Ques 7: Prove that 2-D continuous and discrete Fourier
transforms are linear operations.
Answer:
Understanding Linearity:
• A mathematical operation is called linear if it satisfies the following property:

F(af1 + bf2 ) = aF(f1 ) + bF(f2 )

where a and b are constants, and f1 and f2 are functions.

1. Continuous Fourier Transform:
• For the continuous Fourier transform of a function, we write:
Z Z
F{af1 (x, y) + bf2 (x, y)} = (af1 (x, y) + bf2 (x, y))e−j2π(ux+vy) dx dy

• By the properties of integrals, we can separate the terms:

Z Z Z Z
= af1 (x, y)e−j2π(ux+vy) dx dy + bf2 (x, y)e−j2π(ux+vy) dx dy

• This means:
= aF(f1 (x, y)) + bF(f2 (x, y))

• Thus, the continuous Fourier transform is linear.

2. Discrete Fourier Transform (DFT):
• For DFT, the proof is similar but uses summation instead of integration:
N −1
2πkn
(af1 (n) + bf2 (n))e−j
X
F (k) = N

n=0

• Again, we can separate the terms:

N −1 N −1
−j 2πkn 2πkn
f2 (n)e−j
X X
=a f1 (n)e N +b N

n=0 n=0

• This gives:
= aF1 (k) + bF2 (k)

• Thus, the DFT is also linear.

Conclusion:
• Both the 2-D continuous Fourier transform and the discrete Fourier transform are linear
operations.

• This property allows us to break down complex signals into simpler parts for analysis.

8
Ques 8: What is Digital Image Processing? Discuss some
of its major applications.
Answer:
Digital Image Processing:

1. Digital Image Processing involves using algorithms to perform operations on digital

images.

2. It includes tasks like image enhancement, restoration, and segmentation.

Major Applications:

1. Medical Imaging: Enhances images from X-rays, MRI, and CT scans.

2. Satellite Image Analysis: Used in weather prediction and geographical mapping.

3. Face Recognition: Identifies individuals using facial features.

4. Automotive Industry: Assists in autonomous driving through image recognition.

5. Security Systems: Enhances video surveillance and threat detection.

6. Document Scanning: Improves OCR (Optical Character Recognition).

7. Industrial Inspection: Detects defects in manufacturing processes.

9
Ques 9: Consider two image subsets S1 S2 as shown in
the following figure. For V = 0 determine whether the
regions are: i) 4-Adjacent ii) 8-Adjacent iii) m-Adjacent.
Give reasons for your answer.
Answer:
Determining Adjacency:

1. 4-Adjacent:

• Two regions are considered 4-adjacent if they share a horizontal or vertical neigh-
bor.
• For S1 and S2 to be 4-adjacent, at least one pixel in S1 must be directly adjacent
to a pixel in S2 on the horizontal or vertical axis.
• If any pixel in S1 has a neighboring pixel in S2 with a value of V = 0, then they
are 4-adjacent.
• Example: If S1 contains a pixel at (x, y) and S2 contains a pixel at (x + 1, y), they
are 4-adjacent.

2. 8-Adjacent:

• Two regions are considered 8-adjacent if they share any neighboring pixel, including
diagonal neighbors.
• For S1 and S2 to be 8-adjacent, there must be at least one pixel in S1 that is
adjacent (either horizontally, vertically, or diagonally) to a pixel in S2 with a value
of V = 0.
• This type of adjacency captures all possible neighboring relations, making it more
inclusive than 4-adjacency.
• Example: If S1 has a pixel at (x, y) and S2 has a pixel at (x + 1, y + 1), they are
8-adjacent.

3. m-Adjacent:

• m-adjacency is a generalized form that allows for flexible definitions of adjacency.

• If diagonal connectivity exists, but the pixels involved do not fall into S1 or S2 ,
m-connectivity can be applied.
• This approach helps resolve ambiguities that may arise from using strictly 4 or 8
connectivity.
• The determination of m-adjacency depends on the specific connectivity criteria
defined for the analysis.

4. Determining Adjacency:

• The determination of whether S1 and S2 are 4-adjacent, 8-adjacent, or m-adjacent

will depend on the specific positions of the pixels in the figure and their values.

10
 • For accurate classification, it’s essential to visually assess the connectivity of pixels
and ensure that they meet the criteria defined for each type of adjacency.

5. Conclusion:

• Analyzing pixel connectivity is crucial in image processing tasks such as segmen-

tation and region labeling.
• It influences how images are interpreted and how algorithms are designed to process
them effectively.

11
Ques 10: Find the DFT of f (x) : {0, 1, 2, 1}.
Answer:
Steps to Compute DFT:

1. The Discrete Fourier Transform (DFT) of a sequence is calculated using the formula:
N −1
2πkn
f (n)e−j
X
F (k) = N

n=0

where N is the number of samples in the sequence.

2. Given the sequence f (x) = {0, 1, 2, 1}, we have N = 4.

3. Now, we will calculate F (k) for k = 0, 1, 2, 3:

4. For k = 0:
3 3
−j 2π·0·n
X X
F (0) = f (n)e 4 = f (n) · 1
n=0 n=0

= f (0) + f (1) + f (2) + f (3) = 0 + 1 + 2 + 1 = 4

5. For k = 1:
3 3
2π·1·n πn
f (n)e−j f (n)e−j
X X
F (1) = 4 = 2

n=0 n=0
−j π2 3π
= f (0) · 1 + f (1) · e + f (2) · e−jπ + f (3) · e−j 2

= 0 · 1 + 1 · (−j) + 2 · (−1) + 1 · j = 0 − j − 2 + j = −2

6. For k = 2:
3 3
2π·2·n
f (n)e−j f (n)e−jπn
X X
F (2) = 4 =
n=0 n=0

= f (0) · 1 + f (1) · e−jπ + f (2) · e−j2π + f (3) · e−j3π

= 0 · 1 + 1 · (−1) + 2 · 1 + 1 · (−1) = 0 − 1 + 2 − 1 = 0

7. For k = 3:
3 3
−j 2π·3·n 3πn
f (n)e−j
X X
F (3) = f (n)e 4 = 2

n=0 n=0
−j 3π 9π
= f (0) · 1 + f (1) · e 2 + f (2) · e−j3π + f (3) · e−j 2

= 0 · 1 + 1 · j + 2 · (−1) + 1 · (−j) = 0 + j − 2 − j = −2

8. Final DFT Results:

F (k) = {4, −2, 0, −2}

12
Ques 11: Describe Quantization in short.
Answer:
1. Quantization maps a continuous set of values into a discrete set.

2. Reduces the range of pixel intensity values.

3. Commonly used in image compression techniques.

4. The number of levels affects image quality.

5. Reducing levels introduces quantization error.

6. Applied after sampling to digitize images.

7. Example: Converting 256 gray levels to 16 gray levels.

13
Ques 12: What is Digital Image Processing? Describe in
short.
Answer:
1. Digital Image Processing uses digital computers to process images.

2. Involves operations like filtering, transforming, and segmenting images.

3. Enhances the visual quality of images.

4. Used in various fields like medical imaging, satellite imagery, and computer vision.

5. Allows manipulation of pixel values.

6. Utilizes algorithms for image restoration.

7. Focuses on improving the image’s interpretability.

14
Ques 13: Explain the 4-8 and m connectivity of pixels.
Explain region, edge in context with connectivity of pixels.
Answer:
Connectivity of Pixels:

1. 4-Connectivity:

• In 4-connectivity, a pixel is connected to its immediate horizontal and vertical

neighbors.
• For a pixel located at position (x, y), the 4-connected neighbors are (x + 1, y),
(x − 1, y), (x, y + 1), and (x, y − 1).
• It is commonly used in binary images, where pixels can be either foreground or
background.
• The advantage is that it simplifies the connectivity checks in image processing.
• However, it does not consider diagonal connections, which may lead to fragmen-
tation in the connectivity of shapes.

2. 8-Connectivity:

• In 8-connectivity, a pixel is connected to its immediate horizontal, vertical, and

diagonal neighbors.
• For a pixel at (x, y), the 8-connected neighbors include the same 4 neighbors
from 4-connectivity, plus the diagonal neighbors (x + 1, y + 1), (x + 1, y − 1),
(x − 1, y + 1), and (x − 1, y − 1).
• This approach provides a more comprehensive connectivity analysis, capturing di-
agonal adjacency.
• It is particularly useful in shape analysis and when detecting edges in images.
• However, it may increase the complexity of algorithms due to the larger neighbor-
hood.

3. m-Connectivity:

• m-connectivity is a hybrid approach that allows for flexible connectivity definitions.

• It can include both 4-connectivity and 8-connectivity based on specific criteria to
avoid ambiguities.
• This connectivity model helps maintain consistency in regions that are diagonally
adjacent while providing an option for simpler connectivity.
• It is particularly beneficial in applications where shape connectivity may be am-
biguous.
• The choice between m-connectivity and others can depend on the image content
and processing goals.

Regions and Edges:

1. Region:

15
 • A region is defined as a connected group of pixels that share similar properties,
such as intensity or color.
• It represents a meaningful area in the image that can be processed or analyzed.
• Region properties can include area, perimeter, and shape, which are useful for
image segmentation tasks.
• Connected components are often identified based on connectivity definitions (4, 8,
or m).
• Regions can be extracted using segmentation techniques like thresholding, cluster-
ing, or edge detection.

2. Edge:

• An edge is defined as a boundary between two different regions in an image,

typically where there is a significant change in intensity or color.
• Edges are crucial for identifying object boundaries and shapes within an image.
• They can be detected using various edge detection algorithms, such as the Sobel,
Prewitt, or Canny edge detectors.
• The concept of edge connectivity can be tied to pixel connectivity, determining
how edges are defined in relation to neighboring pixels.
• 4-connected edges may represent sharp transitions along the axes, while 8-connected
edges capture more intricate boundaries, including diagonals.

16
Ques 14: What is the Digital Image Processing?
Answer:
1. It is the manipulation of images using digital technology.

2. Focuses on processing images to improve quality.

3. Involves enhancement, restoration, and analysis.

4. Used in medical, satellite, and surveillance applications.

5. Uses algorithms for noise removal, contrast enhancement, etc.

6. Allows for extracting information from images.

7. Enhances image interpretability.

17
Ques 15: What do you mean by image processing? Ex-
plain the steps of image processing with the help of block
diagram.
Answer:
1. Image Processing: Techniques for manipulating and analyzing images.

2. Image Acquisition: Capturing an image using devices.

3. Image Enhancement: Improving image quality for better visualization.

4. Image Restoration: Recovering an original image from a degraded version.

5. Segmentation: Dividing an image into regions.

6. Feature Extraction: Identifying and measuring key features.

7. Image Analysis: Interpreting processed data.

18
Ques 16: Explain low level, mid level and high level image
processing. Also explain the sampling and quantization
process.
Answer:
Levels of Image Processing:

1. Low-Level Processing:

• Involves basic operations on the image such as noise reduction, image smoothing,
and contrast enhancement.
• These operations are directly applied to the pixel values.
• They focus on improving image quality for further processing.
• Examples include applying filters like Gaussian blur or sharpening to reduce noise.
• The output of low-level processing is generally another image.

2. Mid-Level Processing:

• Focuses on analyzing the image to extract meaningful information.

• It involves operations like segmentation (dividing an image into regions) and feature
extraction (detecting edges or contours).
• Object recognition and shape analysis are common mid-level tasks.
• Example: Identifying boundaries of objects using edge detection techniques like
Canny edge detector.
• The output of mid-level processing is often a set of features or regions extracted
from the image.

3. High-Level Processing:

• Deals with interpreting the information extracted from images.

• It involves understanding the content of an image, like recognizing objects, actions,
or scenes.
• High-level processing is used in computer vision tasks such as facial recognition
and object tracking.
• Example: Analyzing surveillance footage to identify people or activities.
• The output of high-level processing is typically a description or decision about the
image content.

Sampling and Quantization:

1. Sampling:

• Refers to the process of converting a continuous image into a discrete set of pixels.
• The sampling rate determines the number of samples (pixels) per unit area.

19
 • Higher sampling rates result in better spatial resolution, making finer details visible
in the image.
• Lower sampling rates may cause aliasing, where high-frequency details are lost or
misrepresented.
• Example: Sampling a 1024x1024 pixel image provides higher resolution compared
to a 512x512 pixel image.

2. Quantization:

• The process of mapping continuous intensity values into a finite set of discrete
levels.
• It reduces the number of different intensity values (gray levels) that an image can
have.
• Higher quantization levels (e.g., 256 levels for an 8-bit image) result in smoother
gradations between shades.
• Lower quantization levels (e.g., 16 levels) can cause visible artifacts like contouring
or banding.
• Example: Converting a grayscale image from 256 gray levels (8-bit) to 16 gray
levels (4-bit) reduces the image file size but also its smoothness.

3. Effects of Sampling and Quantization:

• Both processes affect the final image quality and file size.
• Higher sampling provides better detail, while higher quantization levels provide
more color depth.
• Reducing either parameter too much can lead to loss of detail or visual artifacts.
• A balance is necessary to maintain image quality while optimizing storage or trans-
mission.

20
Ques 17: Differentiate Correlation and Convolution with
1-D function and a filter example.
Answer:

Correlation Convolution
Measures the similarity between Computes how one function modi-
two signals. fies another.
Not commutative: f ∗ g ̸= g ∗ f . Commutative: f ∗ g = g ∗ f .
No flipping of the signal is involved Involves flipping the filter before ap-
before the operation. plying.
Equation: (f ∗g)(t) = f (k)g(k+ Equation: (f ∗g)(t) = f (k)g(t−
P P

t). k).
Typically used for template match- Commonly used in filtering opera-
ing and pattern recognition. tions and systems analysis.
It measures how well two signals are It gives a modified output signal
aligned. based on the input signal and fil-
ter.
Example: Correlating a signal with Example: Smoothing a signal using
a pattern to find similarities. a moving average filter.

Table 1: Comparison between Correlation and Convolution

21