Image Denoising based on Adaptive BM3D and Singular Value Decomposition

In this work a new version of block-matching and 3D filtering (BM3D) denoising approach introduced by Dabov et al. for denoising the image corrupted by additive white Gassian noise is proposed. The BM3D performs collaborative filtering to the 3D image groups composed by similar image blocks with the fixed hard-thresholding operator. The proposed version of BM3D adopts adaptive block-matching threshold in the block-matching step and the denoising method based on singular value decomposition is used before applying BM3D as the performance of BM3D falls rapidly to strong noise image. To sum up, the proposed method firstly exploits the noise estimation to get the noise level of the given image. Then singular value decomposition is applied to pre-filtering to the high noise level image. Finally BM3D denoising method algorithm with adaptive block-matching thresholds is adopted. Experiment results are given to show that the proposed algorithm achieves better denoising performance than the original BM3D.