Empirical Analysis of Denoising Algorithms for CCTV Face Images

Image denoising is a critical task in image processing that involves the removal of noise or unwanted distortions from an image while preserving its essential features. Most of the commonly captured pictures are obtained using mobile cameras or CCTV surveillance cameras producing video footage of the activities of people who are stationary or in motion. There is a need to restore such captured footage from noise so that it can become evidence for different criminal cases. Denoising face images captured using CCTV is a challenging task due to fine details being affected by noise. In this paper, we evaluate three image denoising techniques Block-Matching and 3D (BM3D), k-Means Singular Value Decomposition (KSVD), and Weighted Nuclear Norm Minimization (WNNM). The performance of these methods is analyzed using Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Visual Information Fidelity (VIF). It is observed that the overall performance of KSVD is better for a Gaussian and Salt and Pepper noise.