Image denoising: who is best?

Abstract

Image denoising is a critical task in image processing, particularly in applications where image quality is crucial. In this paper, we compared the performance of five denoising techniques: TV, NLM, BM3D, DnCNN and FFDNet, on grayscale images corrupted with additive white Gaussian noise (AWGN). The comparison was based on both quantitative and qualitative evaluation of the various methods. The findings revealed that CNN-based methods outperformed the traditional methods significantly, with FFDNet demonstrating better trade-off between denoising performance and computational complexity. Additionally, several directions for future research were discussed.