Research On Image Denoising Based On Non-local Means

Digital image is often corrupted by a variety of noise.Image denoising is a widely-studied and challenging problem and attracts a lot of attention.Nowadays,there are many excellent image denoising methods,in which the non-local means(NLM)method has become a widespread research object for many scholars because of its excellent denoising characteristics.The NLM method can not only effectively remove the noise,but also retain the details of the image.However,it exists some shortcomings,which affect its denoising performance.In this paper,we propose an improved NLM algorithms.Therefore,the NLM method is extended to remove salt and pepper noise by introducing a reference image or noise detection result.Traditional NLM method is lack of accuracy in similarity measurement between pixels,and the selection of filter parameters is not reasonable.This paper proposes an improved NLM algorithm based on method noise.Firstly,the traditional NLM method is used to remove noise from the noisy image,and then the denoising result is smoothed using bilateral filter to extract the residual image signal in the method noise.Finally,the similarity weight function of the NLM is redefined by the combined denoising result and the residual image signal.Moreover,the key filtering parameters is optimized to achieve better denosing results.The traditional NLM method shows poor perfomance in removing salt and pepper noise.In this paper,we based on NLM method to design three kinds of salt and pepper noise removal algorithms.Algorithm one: Adaptive median filter(AMF)based NLM algorithm.Noisy image first is processed using AMF,and then NLM method is applied to the reference image.Algorithm two: Noise adaptive extremum compression median filter(NAECMF)based switching NLM algorithm.We first propose the NAECMF to remove noise,and then,using the switching NLM algorithm improves denoising result further.Algorithm three: Noise detection based NLM algorithm.The proposed algorithm is a two phase denoising method.In the noise detection phase,an effective noise detector is proposed.In the filtering stage,a joint noise template is constructed by using the noise detection result.The template can guarantee the NLM to obtain a reliable similarity weight,and only noise-free pixelsare used for the weighted average.In addition,the iterative filtering scheme is used to remove high-density of noise.