Research On Image Denoising Algorithm Based On Sparse Representation

With the advent of the era of digital media, high-quality images are of importance which can not be ignored among all walks of life. However, during transmission and reception, images are usually inevitably polluted by noise. Thus noise suppression has been the key problem in image processing. Traditional methods can be easily realized, get better performance but can loss details. Recently the development of sparse representation has bring a new idea in denoising method.We discussed denoising method based on sparse representation in this paper. The thesis includes the following three parts:First, the running time of traditional denoising method based on K-SVD is too long although it's denoising quality is higher.In order to reach a good balance between the image denoising and denoising speed. This paper endeavors to improve the denoising algorithm by replacing OMP with the steepest descent ROMP. Meanwhile in this paper we improve the algorithm by eliminating the redundancy.Second, traditional methods can be easily realized but can loss details and lead to edge blurring. To overcome above drawbacks, an image denoising method based on MCA and K-SVD algorithm is carried out in this paper.This paper improved the denoising accuracy by improving the high demand to image sparsity of algorithm.Third, no matter what kinds of denoising methods assume the image can be polluted by gaussian noise only. But the common noise that exists in images is not only gaussian noise but salt&pepper noise.This paper go further analysis on improved image denoising method in resisting mixed noise based on MCA.Compared the proposed algorithm with traditional denoising methods in this paper, and set the algorithm running time and the denoising quality as two index to be analyzed. Simulation results show that our method can get better denoising performance both in PSNR value and visual effects. The running time of the algorithm can be reduced in some extent.But it also isn't the optimal.So the algorithm is still not perfect and need further research.