Research On Block Matching Based Image Denoising And Super-resolution Reconstruction Algorithm

The quality of image is very important because image is one of the main sources for human beings to obtain information. However, there are always some defects on images due to the external environment or image capturing devices. Noise and low resolution are two of many factors to affect our visual perception. This paper is dedicated to suppress image noise, increase the image resolution so as to improve the quality of the image.As far as is known, BM3D, which can remove noise and retain details at the same time, is the best way to denoise. However, it can not yield satisfying results when dealing with impulse noise or mixed noise compared to Gaussian noise. In recent years, the sparse expression and low-rank approximation theory have been widely used in image processing. The joint sparse and low-rank approximation theories yield good results when used to process images with impulse noise especially. Actually, images are often corrupted with Gaussian noise and impulse noise. Most of the existing denoising algorithms can only be used for Gaussian noise or impulse noise, however, when dealing with mixed noise it can not produce satisfying results. This paper proposes a new algorithm that comibines the3D collaborative filtering processing and low-rank matrix recovery, which achieves good results in the treatment for mixed noise.Low resolution is another annoying problem. Image sequence super-resolution reconstruction algorithm can reconstruct high-resolution images using time domain redundancy. However, traditional super-resolution reconstruction algorithm can only deal with the images in the presence of global motion. Elad M introduced non-local block matching into the algorithm of super-resolution reconstruction and gained excellent reconstruction result in dealing with partial motion. Like denoising algorithm based on block matching, the computation of super-resolution reconstruction algorithm based on block matching is also a very time-consuming process. As for this problem, this paper proposes to speed up both of them through GPU programming.