Research On Regularization Reconstruction Of Adaptive Norm Constrained Image

Image restoration included denoising, deblurring, compress sensing (CS) and super-resolution, while these issues are ill-posed inverse problem. Solving the ill-posed inverse problem immediately, which cannot get the ideal solution and much detail information will be destroyed. In this paper, regularization method will be applied to solve ill-posed inverse problem in the image restoration and some works as follows:Firstly, we analyze the regularization method and found that generally chose the data fidelity term which was usually unique. However, many real images include various kinds of noises, which have some localization to deal with the noise. About regularization term, which generally solved the l0 norm. However, l0 norm which is an NP hard optimization problem and very difficult to solve. So generally choosing the l1 norm to solve this problem, which cannot get the ideal sparse solution and difficult to solve the noise heavy tail distribution problem.Secondly, about the present problem and the purpose of improving the quality of the image, an adaptive weighted encoding and l1/2 regularization method is proposed. About data fidelity term, an adaptive norm mixed model method is proposed in this paper, this method can have the advantages of both l1, norm(i.e. edge preservation) and l2 norm (i.e. smoothing characterization).Considering noise distribution change, an efficient adaptive membership degree method is proposed, which can adaptively choose l1, norm and l2 norm to solve the noise problem. Considering many noises have a character of heavy tail and apply an adaptive weighted encoding method, which has a perfectly effect on solving the noise heavy tail distribution problem. Moreover, choosing the regularization term is very important is this paper, l1/2 regularization method is proposed, which can get much sparse solution than l1, regularization operator. Compared with the traditional methods, experimental results demonstrate that the proposed algorithm which has not better reconstruction effect in edge and smoothing region, while reducing iterations and computational cost but verify the proposed method effectively and stability from the subjective and objective.