Research On Image Denoising Based On Non-Local Mean And Sparse Representation Theory

Image denoising is an important part of preprocessing in image processing and recognition, and is an interesting topic for domestic and abroad scholars. In recent years, the effective image denoising methods have been put forward, the most classic representations among them are the theory of the nonlocal mean and sparse representation. This paper focuses on two parts: First is based on the nonlocal average theory of 3D block matching image denoising algorithm, which the block matching caused a large amount of calculation, analysis and improvement, to improve the operation efficiency of the algorithm; Second, for the mixed noise of additive white gauss noise and impulse noise at the same time,based on the sparse representation of mixed noise filtering algorithm has been put forword, and obtained the good experimental effect. In this paper, the main research work is as follows:(1) A BM3 D accelerated algorithm based on integral image is proposed to improve the lower speed problem of the 3D block matching algorithm(BM3D)derived by a large number of block matching. First, Gaussian filter is used to rough process the corrupted image, and then the similarity between image blocks is calculated by combining with integral image diagram. In the second step, integral image can be similarly used by means of converting Wiener filtering into a new form. Experimental results show that the improved algorithm, not only retains the good denoising property of the three-dimensional block matching algorithm, and shortens the operation time by more than 75 percent.(2) For the mixed noise composed of gauss noise and salt and pepper noise, an improved WESNR algorithm has been come up. By analyzing the algorithm performance, using the minimization of the extremum correction at zero point to make it more close to the robust statistics function requirements, and increases the asymptotic reduce random disturbance in the optimization objective function toavoid recursive solving fall into local minimum. Experiments show that the improved algorithm is more effective inhibition of mixed noise, retain more image details.