Research On Image Denoising Based On Non-Local Means

As a fundamental image restoration problem,image denoising has been widely studied during the past decades.The main challenge in image denoising is to suppress noise efficiently while preserving significant image details well.To this end,diverse denoising methods have been proposed,among which non-local self-similarity based denoising methods achieve excellent denoising performance.In this paper,we make an intensive study on non-local means denoising method,and propose some strategies to improve its denoising performance.The main work and innovative points are included as following:(1)Image denoising using rotation invariant non-local means.Original non-local means denoising method focuses only on translational similarity but neglects rotational similarity,which also widely exists in natural images.So,we propose the rotation invariant non-local means denoising method,which makes a full exploitation of image non-local self-similarity.By using rotated versions of the patches,some rotation-similar cases can be measured.Furthermore,we construct the local frequency descriptors,which are rotation invariant and meanwhile robust to noise.So,arbitrary rotation-similar cases are handled.Experimental results demonstrate that the proposed method significantly outperforms the original non-local means denoising method,and obtains better denoising results.(2)Image denoising using quadtree based non-local means with locally adaptive principal component analysis.The non-local means denoising method usually suffers from an unavoidable tradeoff between patch size and the number of potential similar patches.We address this issue and propose the quadtree based non-local means denoising method,which exploits non-local multi-scale self-similarity better.Moreover,by tracking the remaining local noise variance,the locally adaptive principal component analysis is applied to remove the residual noise further.Experimental results demonstrate that the proposed method achieves very competitive denoising performance compared with state-of-the-art denoising methods,even obtaining better visual perception at high noise levels.(3)Image denoising using method noise based shape-adaptive non-local means.In the non-local means denoising method,the calculation of weights can be easily affected by noise,and the square patch shape may limit the number of suitable candidate patches.We address these issues and propose the method noise based shape-adaptive non-local means denoising method,which exploits both non-local self-similarity and local shape adaptation.Based on the denoised image and the residual image in method noise,new weights are calculated,and spatially adaptive patch shapes are defined.Experimental results demonstrate that non-local means with new weights and shape-adaptive patchessignificantly outperforms original one,and achieves competitive denoising performance compared with state-of-the-art denoising methods.