Research On Image Denoising And Texture Completion Based On Non-Local Self-Similarity

Image processing technology is widely used in various fields of modern society.The presence of image noise affects the effectiveness of image processing,so it is necessary to reduce the noise in the image before performing image processing operations.Researchers from both domestic and international communities have conducted in-depth studies on image denoising techniques and proposed various methods for noise reduction.Non-local means filtering,based on the non-local self-similarity of images,has shown remarkable performance and has gained extensive attention from scholars.This paper conducts a comprehensive study of non-local means filtering and explores its application in texture detail completion for images.(1)To address the issue of improper weight allocation in non-local means filtering,an improved weighted non-local means filtering method is proposed.Firstly,bilateral filtering is used as a preprocessing step for the noisy image to enhance the accuracy of weight calculation.Secondly,a relationship function between the weight of the central pixel and the noise value is introduced to adaptively allocate the weight for the central pixel.Finally,outlier detection techniques are employed to eliminate the interference of pixels with low texture similarity on the denoising estimation results.(2)To address the issue of losing texture details while denoising images,a texture completion method based on the residual image is proposed.The residual image consists of the removed noise and texture details after denoising.By extracting the lost texture details from the denoised image and integrating them back,the texture completeness of the denoised image can be improved.Since the signal-to-noise ratio of the residual image is relatively low,a guidance image is used to provide accurate weights for extracting texture details in the denoising process.Firstly,the weight template of mean filtering is improved to preliminarily denoise the noisy image,removing the noise while preserving edge gradient changes.Secondly,the weight function of non-local means filtering is improved by incorporating the correlation judgment of individual pixels,enabling secondary denoising of the noisy image to obtain a guidance image with complete texture details.Finally,the guidance image is used to extract texture details and complete the texture detail loss in the denoised image.(3)Comparative experiments are designed to evaluate the denoising performance of the improved weighted non-local means filtering method.The enhancement effect of the texture completion method based on the residual image on various denoising methods is also tested.Based on the experimental results,the feasibility of the proposed methods is analyzed from both objective data and subjective visual aspects.