Improving Peak Signal-to-Noise Ratio through Wavelet-Based Noise Reduction

Image denoising is one of the basic tasks in image processing, since images may undergo distortions during acquisition, processing, compression, transmission, or reconstruction (Shukla and Patel, no date). Moreover, because both noise and image edges are characterized by high frequencies, conventional denoising often results in unavoidable loss of edge information, producing recovered images that are blurrier or less denoised (Shukla and Patel, no date). This paper presents a comparative study of three filtering approaches — mean filter, median filter, and Wiener filter — versus a proposed wavelet-based soft-thresholding framework implemented in MATLAB (Shukla and Patel, no date). The proposed method applies a discrete wavelet transform to decompose the degraded image into layers, extracts the horizontal, vertical, and diagonal wavelet coefficients, and applies soft thresholding to suppress the noise-dominated high-frequency components while preserving the signal-carrying low-frequency coefficients (Shukla and Patel, no date). The techniques are evaluated on the Cameraman and Barbara test images using peak signal-to-noise ratio. The wavelet technique achieves the highest PSNR on both images — 19.07 on Cameraman versus 14.04 for the Wiener filter and 10.71 for the mean filter, and 10.90 on Barbara versus 09.60 for the mean filter and 08.32 for the Wiener filter (Shukla and Patel, no date). The results demonstrate that the wavelet-based approach offers better performance in preserving the edge structures and detailed textures of the image while improving the PSNR value, making it the best-performing technique among those compared (Shukla and Patel, no date).