The Method Of Image De-noising Based On Multi-dictionary And Sparse Representation

Image may be contaminated noises during each phase in image process applications, e.g.,acquisition, coding and transmission. It significantly decreases the image quality. Thus, image de-noising has always been a hot research topic.Sparse representation has been shown to be more effective than traditional de-noising methods due to adapting to physical environment, less sampling rate and data processing efforts, which significantly enhance the performance of de-noising. In view of this, scholars in related fields have attached great importance to the theory research based on completed atom signal sparse decomposition studies, which has a rapid development in recent years and become a mainstream signal representation. In the field of image de-noising applications, we have made some progress. Firstly, a brief overview is given in this paper on the traditional image de-noising such as classical wavelet de-noising.Then, we explore further the sparse correlation algorithm and techniques based on the initial DCT dictionary.A novel algorithm based on K-SVD and multi-dictionary is proposed. The main contents are as follows:(1)The overview on the image de-noising algorithms and the brief introduction of the over-complete sparse decomposition theory are denoted.(2)We deeply analyzes methods based on orthogonal matching pursuit, Orthogonal Matching, etc. Specifically the complexity and the detail of dictionary updating and signal reconstruction are focused. Furthermore, we describe the dictionary updating algorithm for optimization, named K-SVD.(3)A novel algorithm is proposed based on K-SVD and multi dictionary. This algorithm exploits the initial DCT over-complete dictionary constructed by running K-SVD algorithm on the noisy image and obtains an efficient representation of images, this thesis introduces the idea of multiple-dictionary. Experimental results demonstrate that Compared with the previous algorithms, our proposed algorithm performed better than the traditional denoising algorithm and the sparse representation based denoising algorithm on single DCT redundant dictionary. In addition, our proposed algorithm can preserves more information of original image.(4)The detail of discuss on Analysis K-SVD(Analysis K-means Singular Value Decomposition) algorithm is presented, the experimental results showed its significance.