ScholarWorks@중앙대학교

초록

Despite the advancement of digital imaging technology, noise generated during image acquisition and processing remains a major challenge, degrading image quality and the accuracy of subsequent analysis. This paper systematically reviews the evolution of image denoising techniques, ranging from early statistical filtering to frequency analysis, wavelet transforms, Bayesian models, and the latest deep learning-based approaches. We analyze the statistical characteristics of various noise models, such as Gaussian, salt-and-pepper, and Poisson noise, and conduct a comparative analysis of the performance and limitations of spatial and transform domain filtering, non-local self-similarity-based methods (e.g., NL-means, BM3D), and state-of-the-art models utilizing CNNs, Transformers, and diffusion models. Furthermore, the effectiveness of quantitative performance metrics like PSNR and SSIM is discussed, alongside future research directions emphasizing hybrid models integrating statistical priors with deep learning and self-supervised learning. This review aims to provide essential insights for achieving high-quality imaging and enhancing the reliability of data-driven decision-making systems.

키워드