An Adaptive Edge-Preserving Image Denoising Using Block-Based Singular Value Decomposition in Wavelet Domain

摘要

Image denoising is a quite active research area in the domain of image processing. The essential requirement for a good denoising method is to preserve significant image structures (e.g., edges) after denoising. Wavelet transforms and singular value decomposition (SVD) have been independently used to achieve edge-preserving denoising results for natural images. Numerous denoising algorithms have utilized these two techniques independently. In this paper, a novel technique for edge-preserving image denoising, which combines wavelet transforms and SVD, is proposed. It is adaptive to the inhomogeneous nature of natural images. The multiresolution representation of the corrupted image in wavelet domain is obtained through the application of a discrete wavelet transform to it. A block-SVD based edge-adaptive thresholding scheme which relies on estimation of noise level is employed to reduce the noise contents while preserving significant details of the original version. Comparison of the experimental results with other state-of-the-art methods reveals the fact that the proposed approach achieves very impressive gain in denoising performance.