Researches On Image Reconstruction Based On Spare Redundant Dictionary

Sparse representation can effectively extract the intrinsic information of the signal,which is widely used in signal processing,such as signal denoising,super-resolution,feature extraction and so on.By adding the basis function,the complete basis is extended to over-complete basis,and the better sparsity can be obtained,which is called "sparse dictionary".The compressed sensing theory on the basis of sparse representation,using the sparsity of the signal through the nonlinear reconstruction algorithm,has broken the Shannon-Nyquist sampling theorem and obtained the reconstruction of the signal effectively,greatly alleviated the pressure of high dimensional signal processing,supplemented and improved the signal processing theory.It is of great significance to study the sparse dictionary with excellent performance and improve the reconstruction accuracy.This paper focused on the theory of sparse representation,researched the image reconstruction model on the basis of dictionary learning,starting from the prior knowledge of the image analysis,sparsity and low-rank,this paper proposed two kinds of image reconstruction algorithm,and achieved certain results,as follows:1.In the light of the low of dictionary learning efficiency,learning from the atomic library,this paper exploited entropy to constraint the atomic in order to improve their learning efficiency.Because of the traditional total variation model,it can smooth the image and cause a step effect,this paper utilized a weight isotropic with anisotropic TV regularization.It proposed a new reconstruction algorithm,which combines dictionary learning based on entropy-constraint and a weight total variation.It not only removed noise effectively and robust to noise,but also preserved the texture and detail information better,greatly suppressed the staircase of the Total Variation.2.Images can be decomposed into cartoon and texture,which distinguishly representing the low-rank and sparsity.Considering the image of local and non-local information,and the structure of sparsity,starting from the prior probability,taking advantage of the Gauss mixture model for sparse encoding,this paper proposed an algorithm for image reconstruction of the low rank and Gauss mixed sparse encoding.It can both preserve the sharpness of edges and suppress undesirable artifacts.