MATLAB Image Denoising: Scale 32x32 Filters to 128x128

Overview Steps MATLAB Code Results

Denoising 128x128 Images via Filter Scaling

This technical guide demonstrates how to apply a frequency-domain filter (like a Wiener filter template) originally designed for a 32x32 image onto a 128x128 noisy image using interpolation techniques in MATLAB.

By rescaling the filter, we can apply pre-defined noise reduction characteristics to higher-resolution data, provided we account for coordinate shifting and domain transformations correctly.

Run the Image Processing Simulator

Want to see the Wiener Filter in action with your own parameters?

Launch MATLAB Online Simulator

Step-by-Step Methodology

Step 1: Load or create the noisy image (128x128). We generate a base signal (checkerboard) and add Gaussian noise.

Step 2: Load the Wiener filter coefficients designed for a 32x32 image.

Step 3: Rescale the Wiener filter from 32x32 to 128x128 using bilinear interpolation.

Step 4: Convert the noisy image to the frequency domain using 2D Fast Fourier Transform (FFT2).

Step 5: Convert the scaled filter to the frequency domain. We use ifftshift to ensure the filter origin aligns with the FFT origin.

Step 6: Apply the filter in the frequency domain via element-wise multiplication.

Step 7: Convert the result back to the spatial domain using inverse FFT (IFFT2) and retain the real part.

Step 8: Display the comparative results (Noisy vs. Denoised).

MATLAB Implementation

MATLAB Script

% Step 1: Create a 128x128 test image and add noise original_image = im2double(checkerboard(16, 4, 4)); noisy_image = imnoise(original_image, 'gaussian', 0, 0.01); % Step 2: Create filter coefficients (32x32) filter_32x32 = fspecial('gaussian', [32 32], 2); % Step 3: Rescale filter to 128x128 using bilinear interpolation filter_128x128 = imresize(filter_32x32, [128 128], 'bilinear'); % Step 4: Frequency domain conversion of image noisy_image_fft = fft2(noisy_image); % Step 5: Frequency domain conversion of filter % Centering the filter to prevent image shifting filter_128x128_fft = fft2(ifftshift(filter_128x128)); % Step 6: Apply the filter (Frequency Domain multiplication) result_fft = noisy_image_fft .* filter_128x128_fft; % Step 7: Inverse FFT back to Spatial Domain denoised_image = real(ifft2(result_fft)); % Step 8: Display results figure; subplot(1, 3, 1); imshow(noisy_image, []); title('Noisy Image'); subplot(1, 3, 2); imshow(denoised_image, []); title('Denoised Image'); subplot(1, 3, 3); imshow(filter_128x128, []); title('Wiener Filter Template');

Analysis & Accuracy

To ensure 100% accuracy in the implementation:

• Phase Alignment: Using ifftshift before FFT is critical. Without it, the spatial filter is treated as if its peak is at the top-left corner, causing the denoised image to wrap around the edges.
• Data Types: Images are converted to double to prevent clipping during FFT calculations.
• Domain Logic: The real() function is used after IFFT to remove negligible imaginary components resulting from floating-point errors.

Return to Home →