Research On Denoising The Low Level Image

The technology of digital image processing has rapidly developed in recent years. Nowadays it has been widely used in the field of aerospace, remote sensing satellite and road video monitoring etc., and brought the unprecedented breakthrough to each field. But the quality of digital image is sensitive to the external light intensity. In the case of weak light conditions, the quality of resulting low level image is greatly decreased. It not only has low contrast and fuzzy edges, but also contains strong noise, which cause serious damage to the visual information and bring difficulties to the following image processing work. So the research of denoising low level image has great practical significance. This paper mainly researches the low level image denoising algorithms, the contents are as follows:1. The image denoising algorithm based on the neighborhood estimate the center point by using the surrounding pixels. But this diffusion process will lead to the smoothing image without containing details. To maintain the image texture information, a weight matrix can be introduced to control the diffusion process. So we propose a denoising algorithm based on the diffusion weighted least squares. The aim of denoising can be achieved by minimizing the sum of weighted squared errors in the neighborhood. In order to reduce the effect of edge blurring caused by the diffusion along the gradient direction, we inhibit the larger eigenvalues of weight matrix corresponding to the direction of gradient to control the diffusion process to preserve the edges. In the process of estimate the original image, we introduce the intermediate variable and correlation coefficient. According to the theory of flat area with low gradient value and the edge with high gradient value, we use the gradient information to adaptively adjust the intensity of diffusion to smooth the image. Then we map the correlation coefficient to the constraint space according to the projection theorem and use the steepest descent iterative method to solve it. The simulate experiment prove that the algorithm can effectively remove the noise in the low level image.2. The irrelevant points in the neighborhood would reduce the accuracy of estimation. In addition, there are still many image blocks which are apart from each other but seem similar, and using the redundant information of image can protect the edges. So we provide a novel image denoising algorithm based on a randomly dropout neighborhood and nonlocal similarity. When dealing with the local information, we dropout the neighbor pixels randomly by using the Bernoulli trial, and calculate the corresponding weights of the remaining points based the principle of minimizing the sum of local variance and estimated deviation. When dealing with the nonlocal information, we calculate the coefficient based the similarity structure between the image blocks. At last, we establish a denoising model which combined the local and nolocal similarity and get the solution by using the steepest descent iterative method. The results of experiment indicate that the proposed algorithm can remove the noise and protect the edges.This paper is aimed at denoising the low level image to balance the effect of noise removal and texture reservation. We have obtained some achievements, which have guiding significance for the future low level image denoising research.