Non-local Means Denoising Based On K-L Transform And Grey Relation Degree

Image denoising is an important image preprocessing tool.The purpose of image denoising is to reduce the noise from noisy images and ensure information reliability of the corresponding restored ones.It is the most basic and indispensable steps for image fusion,image reconstruction,image enhancement,image segmentation,feature extraction and analysis.Based on several traditional spatial denoising algorithms,Buades et al.propose non-local means(NLM)creatively.No longer for a single pixel but the image sub-block,NLM opens a new era of spatial domain algorithms.This thesis focuses on NLM and its improvement.The main research contents and achievements are as follows:(1)Based on theoretical knowledge of image denoising,this paper introduces the basic principle of spatial domain algorithms clearly.Simulation experiments are carried out by using peak signal to noise ratio(PSNR)and structural similarity(SSIM)as the evaluation index.Compared with the several classical methods,NLM has higher PSNR value and SSIM value.(2)Traditional NLM algorithm has low robustness because it ignores that different image sub-blocks have different texture features.In order to solve this problem,this paper proposes an improved NLM algorithm which integrates characteristic factor and texture factor into weight calculation.First,the characteristic factor based on K-L transform is constructed and the transformation coefficients are regarded as the feature description of sub-blocks.Then,the texture factor based on grey relation degree is constructed which can well reflect the texture features of sub-blocks.Experimental results demonstrate that the proposed method can effectively remove the image noise and protect the image details.(3)Filter parameters have great dependence on the noise standard deviation.In order to make the whole denoising process be adaptive,this paper proposes an improved block-based method for noise estimation based on K-L transform.With this new method,sub-blocks which belong to the homogeneous region are selected and combined to get more accurate noise estimation.A mount of simulations have been done,and experimental results demonstrate that compared with the traditional block-based method and the improved block-based SVD method,block-based method for noise estimation based on K-L transform has higher accuracy and stability in estimation values.