Study On Non-local Means Algorithms For Image Denoising

As one of the important carrier of information,the image can show abundant and colorful information in a vivid way.The image quality affects people's visual experience and subsequent image processing,but it often deteriorates under noise.Image denoising has been the most basic and critical topic in image processing field and many advanced methods for image denoising have been proposed.This paper focuses on solving the defects of non-local means(NLM)algorithms for image denoising and proposes two improved methods,which are described as follows.The first optimized NLM algorithm is based on fuzzy edge complement,which is aim at solving the defect that image texture cannot be held well using conventional NLM algorithm.In the optimized algorithm,firstly,the image edge information is detected using fuzzy edge complement method and the edge feature map is obtained.Then,the similarity of blocks in the edge feature map is used to modify the similarity of blocks in the noise image.Next,the information of structure in image blocks is analyzed quantitatively based on the edge feature map.Finally,the similar weight parameters are adaptively selected based on the volume of structure information and noise level.Experiment results show that the optimized NLM algorithm can achieve better effect on holding the detailed texture features.The second optimized NLM algorithm is based on median filter,which is aimed at solving the defect that salt-and-pepper noise cannot be removed effectively by conventional NLM algorithm.In the optimized algorithm,firstly,the pixels of salt-and-pepper noise are detected by the impulse detection.Then,according to the detection results,a pre-processed denoising image can be obtained by median filter.Next,the similar weights in the NLM algorithm are modified according to the characteristics of salt-and-pepper noise.Finally,the pre-processed denoising image is further denoised using the modified non-local mean algorithm.Experiment results show that the method can still achieve satisfactory denoising effects even if the noise level is high.