Image De-noising Based On Sparse Representation And Trained Dictionary

Image de-noising is one of the most important issues in the digital image processing.Sparse representation and learned over-completed dictionary have attracted researchers’attention recently with the development of compressed sensing theory. Therefore, imagedenoising based on image sparse representation and over-completed dictionary becomesone of the frontier issues in the field.This paper deeply researches on sparse representation which based on compressedsensing theory, compared and summarized pursuit algorithms for image reconstruction.The image has sparse representation based on over-completed dictionary, and the sparserepresentation may represent efficiently the geometrical characteristics of the images. Soin the dictionary construction, we using method of optimal direction to training thedictionary in one transform domain. The MOD algorithm is very slow ,because it has theinverse process. So ,we use singular value decomposition algorithm to training thedictionary. And compared the results of the image de-noising of the two method.Considering we can using the over-completed dictionary in two orthogonal bases, wepropose a new image de-noising approach based on learned dictionary with union of twoorthogonal bases. In the dictionary construction, the dictionary consists of the twoorthogonal bases. The efficient Block Coordinate Relaxation algorithm can be used todictionary train and the Singular Value Decomposition is followed to update the onechosen basis effectively. Experiment results show that the proposed method can fulfill theimage de-noising effectively.