Research On Image Denoising Method Based On Sparse Representation Of Non-local Similar Structure Group

As an important medium for transmitting information,images have penetrated into all aspects of people's lives.However,due to various factors such as imaging equipment and external environment,the collected images are often interfered by noise,which not only brings challenges to the subsequent image processing,but also seriously affects the human visual experience.Therefore,denoising preprocessing of images has always been a research hotspot.Although the traditional denoising algorithms are relatively easy to implement,they're not ideal for edge-preserving and artifacts suppression.In recent years,the rise of sparse representation theory and the widespread application of Nonlocal Self-similarity have provided new ideas for image denoising.In this thesis,we take a research on the natural image denoising problem combining with the above two methods and basing on the nonlocal similar structure group sparse representation.The specific work can be listed as follows(1)When denoising with sparse representation,obtaining a more precise sparse coefficient usually means higher image restoration quality.The obtained coefficients are often inaccurate because of noise interference.In order to solve this problem,this thesis constructs a double/1-norm optimization model of group sparse representation,in which the/1-norm regularization term and the sparse residual regularization term are combined to constrain the group sparse coefficient,so as to obtain more accurate sparse coefficient.In addition,considering the great influence of regularization parameters on the denoising effect,a method of adaptive regularization parameters is designed with Bayesian formula to further improve the denoising performance(2)Since the above model only considers the prior knowledge of the degraded image,which will cause inaccuracy of reconstructed images due to the influence of noise.Therefore we consider Nonlocal Self-similarity priori into both the degenerate and external natural images,and the approximate estimation value of the original image group sparse coefficient in the sparse residual term is obtained by establishing a Gaussian Mixture Model and nonlocal weightingFinally,adding Additive White Gaussian Noise,the international standard dataset is applied to verify the performance of our denoising algorithm of the thesis proposed.And compared with the traditional denoising algorithms from two aspects of objective and subjective evaluation.The experiment results show that the denoising algorithm of this the could remove noise efficiently.At the same time,artifacts can be effectively suppressed and the details of the image can be preserved adequately.