Research On Improved Non-local Means Image Denoising Algorithm

With the rapid development of computer technology,the demand of digital image quality in various fields has been improved.However,in the process of acquisition and transmission,due to the influence of external environment and other factors,the image will inevitably be disturbed by various kinds of noise,which leads to the deterioration of image quality and serious impacts on the image reading and subsequent processing.Non-Local Means(NLM)denoising algorithm is a better filtering algorithm proposed in recent years,but there are still some problems such as noise interference to similar information and serious loss of image information.Therefore,this paper focuses on the similarity measurement and weight allocation in NLM denoising algorithm,and proposes three improved methods to improve the denoising performance of the algorithm.Research work and contribution of this paper:In order to improve the utilization of high similarity pixels in denoising estimation,a non-local means denoising algorithm based on weight redistribution is proposed.In this method,the ratio between the maximum weight and the mean value in the search neighborhood is used to judge the similarity degree between the image blocks in the search neighborhood and the current neighborhood,and the proportion of the similar pixel points in the pixel estimation is adjusted according to the similarity degree,so as to increases the utilization of high-similar pixels.A non-local means denoising algorithm based on edge detection and multi-feature fusion is proposed.The improved Sobel operator is used to extract the edge structure of the image,and different similarity measures are used in edge and flat region according to the distribution characteristics of pixels.Meanwhile,the similarity neighborhood is searched to the maximum degree to improve the search rate of similar neighborhood,which is used to reduce the influence of noise on similar pixel search in the original NLM(NLM)algorithm.To solve the problem that the presence of noise leads to interference of dissimilar pixels when the NLM algorithm is used to measure the similarity of pixels,a dual kernel non-local means denoising algorithm based on gradient feature is proposed.The Euclidean distance between pixels and the gradient feature are used to measure the similarity of the neighborhood,and a new kernel calculation method is used to replace the exponential kernel in the NLM algorithm to obtain more effective weight distribution.In this paper,a large number of experiments are carried out on the proposed algorithm by using images with different degrees of Gaussian noise.The experiment results show that the proposed improved method can reflect the similarity between the neighbors accurately when the proposed filtering algorithm compared with the NLM algorithm and the improved NLM algorithm with Gaussian kernel and Wave kernel,respectively.As the same time,the noise is removed and the edge texture information is retained,thus the denoising effect is improved significantly.