An Effective Aggregation Function for Image Denoising Based on Low Rank Matrix Completion

In this paper, an aggregation function is proposed to improve the performance of the conventional denoising method based on low rank matrix completion. Since this method determines the denoised value of each pixel by averaging the corresponding pixels in the denoised image patches, the performance can be improved by a reasonable aggregation function. The proposed aggregation function exploits the intensity similarity and geometry closeness of the denoised patches, to reduce the unwanted artifacts in the synthesized denoised image. Experimental results show that the proposed method achieves substantial PSNR improvement as compared with the conventional denoising algorithm.