Research On Image Restoration Technology Based On Sparse Model

Image restoration is a process that use some way to make the degraded image become the high quality image. It is a hot issue and Basic research in the field of image processing, the research of image restoration technology is of great significance and far-reaching influence. Image restoration via sparse representation have a good recovery effect and strong robustness. In this paper, we mainly focus on the application of sparse representation in image restoration, and improve the existing algorithms. The main work is as follows.Describe the theory of image restoration technology and sparse representation. Firstly, the basic principle of image restoration is described, and introduce the mathematical model of image degradation process and recovery process,Secondly, the basic principle of sparse representation is introduced. Finally, analyze the image quality evaluation method used in this paper.Atmospheric turbulence is one of the important forms of the irregular movement of the atmosphere as well as the molecular motion. It can make the image blurred, and seriously affect the recognition and tracking of the observation target. In this paper, the restoration of atmospheric turbulence degraded image is studied. First analysis the specific characteristics of the atmospheric turbulence degraded image, combined with the theory of sparse representation, fully explain the dictionary learning can restore the atmospheric turbulence degraded image; then use the DCT over complete dictionary, the K-SVD Dictionary of global and adaptive Dictionary to restore the atmospheric turbulence degraded image, and take the traditional Wiener filtering algorithm as contrast experiment, make the corresponding experiments and data analysis. The simulation results show that the sparse representation is effective in the restoration of atmospheric turbulence degraded images, and the results show that the image obtained by the method of dictionary learning is more effective.In order to reduce the computational complexity, improve the speed of the algorithm, solving the ill posed problems of image restoration. Combine the sparse regularization and non local self similarity to restore image,it can constraints on the image by two effective ways, it will be the ill posed problem which is difficult to solve turn to the posed problems which is easily to solve. Firstly, a brief analysis is made on sparse regularized restoration image. Through the basic concepts of this sparse regularization indicate the necessity use of non local self similarity theory. According to the randomness of the k-mean algorithm, using the new improved method in clustering, which can effectively improve the clustering accuracy. Use the K-svd algorithm and the BM3 D to make contrast experiment. Through the analysis of experimental data, compared with the other two algorithm, this algorithm let the image have better retention of the structure information, get better visual effect, and the calculation complexity reduced.