Digital image noise removal based on collaborative filtering approach and singular value decomposition

Image denoising is a crucial step in order to improve digital image quality. Furthermore, the digital image in sparse format especially in low-rank structure has been utilized in several multimedia applications. Non-local similarity algorithm is used to increase the level of noise removal methods and improve the image visual quality due to the pivotal correlation in the image inter-patches, and intra-correlation with a low-rank prior of the texture itself where the second generation wavelet is exploited to develop a similar coefficients to the image matrix. In order to solve the burden of selected contaminated pixels in the sub-patches, the singular value shrinkage is used to guarantee the elimination of high noisy pixels. Furthermore, the proposed method uses random matrix in order to practically choose the level of singular value threshold. As the experimental results depicts, the proposed algorithm has superior performance in peak signal to noise ratio (PSNR), structural similarity index (SSIM), image quality assessment (IQA) and figure of merit (FOM) in comparison with state of the art noise removal techniques and can recover better details and less artifacts and blurring. In addition, to show the quality of complexity time of the proposed method, execution time of the proposed method and several denoising methods has been examined.