BM3D optical image denoising algorithm based on SVD noise estimation

Abstract

Image denoising represents a crucial technique for recovering visual information, whereby irrelevant noise is effectively suppressed, thus enhancing the quality and recognition of the image. It has been demonstrated that existing image-denoising algorithms can achieve considerable denoising effects when the noise intensity is known. However, natural images are affected by multiple unknown noise sources during the generation and transmission processes. The unpredictable nature of the noise intensity often results in unsatisfactory image-denoising effects. Therefore, this paper proposes a block-matching three-dimensional cooperative filtering (BM3D) optical image denoising algorithm based on singular value decomposition (SVD) noise estimation. First, the optical image is decomposed by SVD. Then, using the tail of singular values to estimate the noise variance of the optical image. Subsequently, the noise estimated value is employed as prior information for BM3D, facilitating the denoising analysis of the optical image. The final step is to use the PSNR to evaluate the algorithm's effectiveness. Experimental results show that compared with the classical image denoising algorithm, the denoising effect of the algorithm in this paper can better preserve image details, reduce noise residue, and perform better in terms of visual effects and PSNR.