The Application Of Sparse Representation In Image Denoising

Image denoising is a classical and difficult problem in image processing, so it is a hot topic. Denoising quality has a direct impact on the quality of subsequent image processing. To make the denoising algorithm with higer SNR and better visual effect, the research is conducted through the following two sparse representation methods:1) The traditional threshold denoising algorithm such as the Block translation invariant stein threshold (B-stein), soft threshold and hard threshold denoising methods can obtain good denoising effect, but open questions like pseudo phenomenon and fuzzy edge appear in restored image. Therefore, an Adaptive Block Threshod based on Translation Invariance (ABTT1) algorithm is proposed for image denoising in this paper. The wavelet coeffiicents of the image is subdivided into sub-block with a new method, and the sub-block threshold function are established to compute the threshold based on the information of sub-block adaptively. Moreover, the property of shif invariant is used by the proposed algorithm. The experimental results show that Comparing with the traditional B-stein, soft threshold, hard threshold denoising method. ABTTI algorithm not only obtains higher SNR and computing efficiency, but also has good performance in eliminating artifacts.2) The image attained from wavelet threshold denoising is not the optimal solution of the original image, and the threshold of sub-block results in computation inefficiency. In order to solve these problems, the image denoising based on the variation method is introduced on the basis of the wavelet threshold denoising. It is a kind of new method with huge potential. This paper systematically studies the basic theory of the total variation denoising model, transforms it into a minimum problem of an energy function and uses Euler equation to study the solution of the problem. TV and Soblev regularization of the two classical total variation denoising model are solved with a gradient descent method. In a addition, we verify the influence of the model parameters on the denoising performance, such as SNR and computational efficiency. The gradient descent uses iterative calculation to get the optimal image approximate the original one. Compared with the wavelet threshold denoising method, total variation denoising can obtain higher SNR and computational efficiency.