DIGITAL IMAGE PROCEESSING

Batch-16

T3-ASSIGNMENT

211FA04096 211FA04170 211FA04257 211FA04573

SEC-B SEC-B SEC-B SEC-B

Abstract— a) Contrast various types of noise [Link] Noise:

and their causes, which can corrupt an image? Cause: Although not practically present, uniform
noise is often used in numerical simulations to
Also describe their mathematical formulations. analyse systems.
Mathematical Formulation:
b) Design an algorithm to add the noise signal The PDF of uniform noise is given by:
(α [m,n] = cos (60πm/M ) cos(120πn/N) ) to 1
your image, where M and N corresponds to the P(x )=
b−a
dimensions of the image you have chosen. where:
x represents the noise value.
c) Design an algorithm for Indicate the a and b define the range of uniform distribution.
frequency domain approach to remove this 4. Impulse (Salt-and-Pepper) Noise:
noise through various filters. Analyze which Cause: Impulse noise results from quick transients,
such as faulty switching in cameras.
frequency domain filter is best suitable for Mathematical Formulation:
noise removal.
DESCRIPTION
[Link] Noise:
Cause: Gaussian noise arises due to factors such as
electronic circuit noise and sensor noise caused by
poor illumination or high temperature.
Mathematical Formulation:

Where: 5. Quantization Noise:

N(x, y) is the Gaussian noise at pixel location (x, y). Cause: Quantization noise arises from the
μ is the mean of the Gaussian distribution, quantization process when analog signals are
representing the average intensity of the noise. converted into digital format.
σ is the standard deviation of the Gaussian Mathematical Formulation:
distribution, controlling the spread or amplitude of Q= (Q/2) × U(−1,1)
the noise. Where:
N(0,1) is a random sample from a standard normal Q represents the quantization step size, which is the
distribution with mean 0 and standard deviation 1. difference between adjacent quantization levels.
[Link] Noise: U(−1,1) is a uniformly distributed random variable
Cause: Exponential noise is commonly present in ranging from -1 to 1.
cases of laser imaging.
Mathematical Formulation: ALOGRITHM(B)
The PDF of exponential noise is given by: 1. Read the original image:- The code reads an
−ax image file named `'[Link]'` and stores
P ( x )=a e it in the variable `original_image`.
where: 2. Get the dimensions of the original image:-
x represents the noise value. The dimensions of the original image are
a is a constant obtained using the `size` function. The
 variable `M` represents the number of rows
(height) and `N` represents the number of
columns (width).
3. Generate the noise signal:- Two-dimensional
grids `m` and `n` are created using
`meshgrid` representing the spatial sOUTPUT(B)
dimensions of the image. A noise signal
`alpha` is generated based on the equation
`alpha = cos(60*pi*n/N) .*
cos(120*pi*m/M)`.
4. Normalize the noise signal:- The noise
signal `alpha` is normalized to the range [0,
1] by subtracting the minimum value and
dividing by the maximum value.
5. Scale and convert the noise signal to uint8:-
The normalized noise signal is scaled to the
range [0, 255] and converted to the `uint8`
data type to match the image data type.
6. 6. Add the noise signal to the original
image:- The noise signal `alpha_scaled` is
added to the original image `original_image`
element-wise to create a noisy image
`noisy_image`.
7. 7. Display the original image and the noise- ALGORITHM
added image:- The original image and the 1. Input:
noisy image are displayed side by side using - Read the noisy image (`[Link]`).
the `subplot` function with appropriate titles. - Convert the image to grayscale if it's in color.
2. Preprocessing:
CODE(B) - Convert the grayscale image to double precision
original_image = imread('[Link]’); for numerical processing.
[M, N, ~] = size(original_image); - Determine the dimensions (`M` x `N`) of the
[m, n] = meshgrid(1:N, 1:M); grayscale image.
alpha = cos(60*pi*n/N) .* cos(120*pi*m/M); 3. Generate Noise Signal:
alpha = alpha - min(alpha(:)); - Create a noise signal `alpha` using cosine
alpha = alpha / max(alpha(:)); functions of varying frequencies.
alpha_scaled = uint8(255 * alpha); - Add the noise signal to the original grayscale
noisy_image = original_image + alpha_scaled; image to generate the noisy image.
subplot(1, 2, 1); 4. Fourier Transform:
imshow(original_image); - Perform Fourier Transform on the noisy image.
title('Original Image'); - Shift the zero frequency component to the center
subplot(1, 2, 2); of the spectrum.
imshow(noisy_image); 5. Define Filters:- Define the parameters for different
title('Image with Added Noise'); filters:
- `sigma_low`: Standard deviation for low-pass
filter.
- `sigma_high`: Standard deviation for high-pass
filter.
- `sigma_bandpass_high`: Standard deviation for
the high-pass Gaussian filter.
- `sigma_bandpass_low`: Standard deviation for
the low-pass Gaussian filter.
- Create the following filters:
- Low-pass filter using Gaussian function.
- High-pass filter using Gaussian function with
zero-sum correction.
 - Band-pass filter by subtracting low-pass from lowpass_filtered_F = F .*
high-pass filter. fftshift(fft2(lowpass_filter));
6. Apply Filters in Frequency Domain: highpass_filtered_F = F .*
- Multiply the Fourier-transformed noisy image fftshift(fft2(highpass_filter));
with each filter in the frequency domain to obtain bandpass_filtered_F = F .*
filtered images. fftshift(fft2(bandpass_filter));
7. Inverse Fourier Transform: lowpass_filtered_image =
- Perform Inverse Fourier Transform on each ifft2(ifftshift(lowpass_filtered_F));
filtered image to obtain spatial domain highpass_filtered_image =
representations. ifft2(ifftshift(highpass_filtered_F));
8. PSNR Calculation: bandpass_filtered_image =
- Calculate PSNR for each filtered image compared ifft2(ifftshift(bandpass_filtered_F));
to the original grayscale image using `psnr` function psnr_lowpass = psnr(uint8(image_gray),
from MATLAB's Image Processing Toolbox. uint8(abs(lowpass_filtered_image)));
9. Output PSNR Values: psnr_highpass = psnr(uint8(image_gray),
- Display PSNR values for each filtered image. uint8(abs(highpass_filtered_image)));
10. Display Results: psnr_bandpass = psnr(uint8(image_gray),
- Display the original grayscale image, noisy uint8(abs(bandpass_filtered_image)));
image, and the three filtered images (low-pass, high- fprintf('PSNR for Low-pass Filtered Image: %.2f\n',
pass, and band-pass) in separate subplots. psnr_lowpass);
fprintf('PSNR for High-pass Filtered Image: %.2f\n',
psnr_highpass);
fprintf('PSNR for Band-pass Filtered Image: %.2f\n',
SOURCE CODE psnr_bandpass);
% Display original and filtered images
image = imread('[Link]’); figure;
if size(image, 3) == 3 subplot(3, 2, 1), imshow(uint8(image_gray)),
image_gray = rgb2gray(image); title('Original Image');
else subplot(3, 2, 3), imshow(uint8(noisy_image)),
image_gray = image; title('Noisy Image');
end subplot(3, 2, 4),
image_gray_double = double(image_gray); imshow(uint8(abs(lowpass_filtered_image))),
[M, N] = size(image_gray_double); % Get title('Low-pass Filtered Image');
dimensions of the original image subplot(3, 2, 5),
alpha = cos(60*pi*(1:M)/M)' * cos(120*pi*(1:N)/N); imshow(uint8(abs(highpass_filtered_image))),
noisy_image = image_gray_double + alpha; % Add title('High-pass Filtered Image');
noise to the original image subplot(3,2,6),
F = fftshift(fft2(noisy_image)); % Shift zero imshow(uint8(abs(bandpass_filtered_image))),
frequency to center title('Band-pass Filtered Image');
% Define filters in the frequency domain
sigma_low = 10; OUTPUT
sigma_high = 50;
sigma_bandpass_high = 200;
sigma_bandpass_low = 50;
lowpass_filter = fspecial('gaussian', [M, N],
sigma_low);
highpass_filter = fspecial('gaussian', [M, N],
sigma_high);
highpass_filter = highpass_filter -
mean(highpass_filter(:)); % Ensure zero-sum for
high-pass effect
bandpass_filter = fspecial('gaussian', [M, N],
sigma_bandpass_high) - fspecial('gaussian', [M, N],
sigma_bandpass_low);
% Apply filters in the frequency domain