Research On Sparse Representation Denoising Algorithms

Signal sparse representation is an emerging signal analysis and synthesis method, itsaim is that the over-complete dictionary uses as few atoms as possible to represent signal.Sparse representation is widely used and has been successfully applied in the field of imagede-noising. During the process of sparse representation the noise can be filtered out. Sparserepresentation de-noising methods can be divided into two methods based on local andnon-local approaches. K-SVD is a local de-noising method. Besides, both NLM and BM3Dare non-local de-noising methods. In this paper, we propose the sparse representationde-noising method based on local and non-local de-noising method.For local de-noising algorithm, K-SVD has a better de-noising treatment effect onstrong texture image than strong structural image. So we propose a new de-noising method,morphological component analysis de-noising which divides the image into structural layerand texture layer. The structural layer is de-noised by non-local BM3D algorithm and thetexture layer is de-noised by K-SVD algorithm. At last we combine the two de-noisedpictures as the result. Compared with single K-SVD and BM3D algorithms, the de-noisingability has been dramatically improved.For non-adjacent but structure similar image blocks are redundant, an algorithm thatclustering algorithm introduced into sparse model is proposed. The dictionary structured byPCA and the cluster centers are updated. At last, the result shows that the proposed methodis superior to local de-noising algorithm K-SVD.