Research Of Image Denoising Based On Nolocal Similarity And Spare Representation

Image denois ing is to obtain the origina l image of the real scene based on the prior knowledge and the degrada tion model of the image itself. It is an important means to ensure that people can understand the image informat ion. According to different source of no ise and pollut ion models, noise can be divided into various forms. In this paper,we mainly study the mixed noise pollut ion of t he Gaussian no ise and impulse noise. The sparsity and nonlocal similarity are two important prior knowledge of the image.In this paper,we comb ine the two kinds of knowledge with t he variationa l modal,and then get the denoised image by solving the variational model. The main work is as follows:1?We analyze the model of Gauss ian no ise ?impluse noise and the mixed noise.Study on simulat ion of s uper denois ing algor ithm previous ly proposed,and then analyze the advantages and disadvantages of these algorithms.2?The princip le, model and algorit hm of image denoising based on sparse representation are studied in detail. The sparse representation is ap plied to the remova l of mixed noise,and we get the prior knowledge of sparse representation of the image by construcing an effective training dictionary,and then get the denoising image by solving sparse model.3?A study of get the nonlocal similarit y of the image under the mixed noise is studied. The experimenta l results show that the use of non local similarit y for mixed noise removal has a better effect.4?An analys is is made on how to merge the sparse representation and nonlocal similar ity. This two prior knowledge are integrated into the regular izat ion of the variationa l model,and then optimizat ion design for data fidelity term and regularizat ion.Through solution the variationa l model can obtain the sparse representation coefficie nt and then get the ideal denoising image.