IMAGE DENOISING USING LOW RANK MATRIX COMPLETION VIA BILINEAR GENERALIZED APPROXIMATE MESSAGE PASSING

摘要

A robust denoising algorithm, which is capable of removing mixed noise from natural images, is proposed based on the bilinear generalized approximate message passing (BiGAMP). This algorithm utilizes the non-local self-similarities of natural images, changes the problem of removing mixed noise to a low-rank matrix completion problem, and takes use of BiGAMP to solve this inherently bilinear problem. Experimental results show that this BiGAMP-based image denoising algorithm is good at removing Gaussian noise mixed with impulsive noise. Its performance is better than the well-known block-matching and 3D filtering (BM3D) algorithm and some state-of-the-art image denoising algorithms via low-rank matrix completion.