A Block-Grouping Method for Image Denoising by Block Matching and 3-D Transform Filtering

Abstract

Image denoising by block matching and threedimensional<br />transform filtering (BM3D) is a two steps state-ofthe-<br />art algorithm that uses the redundancy of similar blocks in<br />noisy image for removing noise. Similar blocks which can have<br />some overlap are found by a block matching method and grouped<br />to make 3-D blocks for 3-D transform filtering. In this paper we<br />propose a new block grouping algorithm in the first step of<br />BM3D that improves the performance of denoising algorithm<br />especially in heavy noise conditions. In heavy noise conditions,<br />BM3D causes some artifacts in the filtered image. These artifacts<br />are reduced by the proposed block grouping algorithm. In the<br />proposed block grouping method, beside of a similarity measure<br />used for block matching, the amount of overlap between blocks is<br />considered. Experimental results show that the proposed block<br />grouping method can improve the performance of BM3D in<br />terms of both peak signal-to-noise ratio (PSNR) and visual<br />quality.