A modified block matching 3D filtering by using variable block sizes

Abstract

We propose an improved image denoising method based on the block matching 3D filtering (BM3D) by using variable block sizes. The previous BM3D mainly consists of two steps: a basic estimate by hard thresholding and a final estimate by Wiener filtering. In order to exploit the spatial correlation among adjacent pixels in an image, we specifically employed the concept of the variable block sizes in the block matching process. The block division simply depends on the variance and the mean, which are key parameters to describe the histogram of the image. Therefore, the proposed algorithm efficiently reduced the noise in the large and plain areas and preserved the fine details at the same time.