MATLAB Implementation of Wavelet Denoising for Images

Wavelet denoising is a common image processing technique particularly effective for removing Gaussian noise. The MATLAB implementation typically involves three key steps: adding noise, performing wavelet-based denoising, and evaluating the denoising results.

First, we need to add Gaussian noise to a grayscale image. Gaussian noise is characterized by its normal distribution properties, where noise intensity can be controlled by setting the mean and variance parameters. In MATLAB, this can be implemented using built-in random number generation functions like randn() combined with image intensity scaling to achieve the desired noise level.

Next, wavelet transform is applied to the noisy image for denoising. The core principle involves decomposing the image into different frequency subbands through wavelet decomposition, producing approximation coefficients (low-frequency components) and detail coefficients (high-frequency components). Since noise primarily resides in high-frequency regions, we can suppress it by applying thresholding techniques (soft thresholding or hard thresholding) to the wavelet coefficients. MATLAB's Wavelet Toolbox provides comprehensive functions such as wavedec2() for 2D wavelet decomposition and waverec2() for reconstruction, enabling multi-level wavelet analysis with various wavelet families like 'db4' or 'sym8'.

Finally, we calculate the Peak Signal-to-Noise Ratio (PSNR) to evaluate denoising performance. PSNR is a widely used image quality metric where higher values indicate better denoising results. The calculation involves computing the Mean Squared Error (MSE) between the original and denoised images, then converting it to a logarithmic signal-to-noise ratio using the formula: PSNR = 10*log10(MAX^2/MSE), where MAX represents the maximum possible pixel value.

The entire workflow demonstrates the advantages of wavelet denoising in image restoration: effectively removing noise while preserving edge information and fine details. By adjusting parameters such as wavelet basis functions, decomposition levels, and thresholding strategies (using functions like wthresh()), users can further optimize denoising performance for specific applications.