The Study Of Image Denoising Methods Based On The Non-local Means

Due to interference from the external and internal environment,the noise is inevitable in the process of image acquisition and transmission,which affects the understanding of the transmitted image information seriously.Therefore,the denoising of transmitted images has been one of the important researches in image processing,and has attracted more and more attention.Non-local mean filtering algorithm(NLM)is one of the effective denoising algorithms.To obtain the estimated grayscale values of image pixels,NLM needs to calculate the weighted average of the grayscale values of pixels in all approximate fields in the image by using the redundant information in the image.In this paper,how to determine the weighted kernel function and the measurement function of similarity becomes the most critical problem.Baudes et al.used the european distance based on Gaussian weighting to measure the similarity between pixel domains and choose exponential functions as weighted kernel functions.However,the improved algorithm only has a good denoising effect in the area where the image is relatively smooth,but the denoising ability is weak in the area such as edge or texture.Hence,the Gaussian weighted distance has certain limitations in determining the weight coefficient.this paper improves the two aspects of weighted kernel function and similarity measurement.In order to solve the problem of unequal distribution of weights in exponential weighted kernel function,a more appropriate weight allocation function is proposed,which can allocate higher weights to places with high similarities,while allocating smaller weights to places with low similarities.At the same time,the algorithm also ensures that weight values attenuate rapidly in excessive regions from high similarity to low similarity.To obtain a more appropriate similarity measurement function,a new similarity judgment function,which combines structural similarity algorithm and European distance,is proposed.After the denoising experiment of noise image,the experimental results show that the new algorithm not only improves the denoising ability of non-local mean filtering significantly,but also better preserves the original information of the image in the edge texture part.