Research On Dongba Painting Denoising Based On Non-local Mean And Directional Mean

Dongba painting originated from the sacrificial activities of the Dongba religion of the ancient Naxi people in southwest China.It not only witnessed the history of Dongba religion,but also recorded the history of the integration of the Naxi people with other ethnic groups.It is the basis for the study of Dongba culture.However,due to various reasons such as drawing skills,painting materials,digitization,etc.,Dongba painting have problems such as noise,messy textures,and uneven colors.Image denoising technology is based on a certain noise model,such as Gaussian white noise,to estimate the original image and restore the original color,structure and texture of the image.Because Dongba painting have a rich variety of colors and lines of different size are crisscrossed,existing image denoising algorithms cannot effectively remove messy textures and repair color unevenness when they are processed directly on Dongba painting.In order to inherit Dongba painting and restore the original colors and textures of Dongba painting,this article will take Dongba painting as the research object for image denoising research.With the continuous development of image denoising algorithms,the spatial filtering method based on the non-local mean has achieved better results in the denoising process,among them,BM3 D algorithm has become the benchmark algorithm for evaluating the denoising effect because of its easy implementation and obvious denoising effect.However,by studying the BM3 D algorithm,it can be found that:(1)It only matches similar image blocks by calculating the similarity between image blocks in the original noise image,and the noise in the noise image has a greater impact on the similarity calculation,which may cause the similar block group obtained has a large error.(2)It can handle structured textures composed of lines well,but it cannot handle unstructured textures,such as uneven color in Dongba painting images.To this finish,this thesis proposes a non-local denoising framework(SVC-NL)supported singular value channel associate degreed an edge-guided directional mean smoothing procedure(DMS)respectively.The main work of this thesis is as follows:1.Proposed a non-local denoising algorithm based on singular value channel(SVC-NL).(1)Aiming at the problem that processing directly on noisy images with higher noise levels will lead to poor algorithm robustness,this thesis proposes an iterative singular value decomposition.The image is reconstructed iteratively based on the percentile of the singular value,the singular value is used to decompose the main information and noise in the image,and a reconstructed image with low noise level is obtained.(2)When grouping image blocks in a noisy image with a high noise level,the noise will cause a larger error in calculating the similarity.Therefore,this thesis looks for similar blocks in the reconstructed image with a lower noise level to reduce the calculation error.(3)In iterative singular value decomposition,when the percentile of the singular value is less than 100,the reconstructed image will also lose part of the image information while removing noise,which may affect the final denoising effect.Therefore,after obtaining the similar block group in the reconstructed image,this thesis uses its position information to match the image block at the corresponding position in the noise image,and then performs a weighted average of the two to reduce the influence of information loss on the denoising effect.2.Proposed an edge-guided directional mean smoothing algorithm(DMS).According to the uneven color problems in Dongba painting,based on the obvious colors and lines of Dongba painting as a whole,the color difference formula is used to quantify the color difference between the pixel values of different colors,and the edge is extracted by counting the local color difference.Then an edge-guided directional mean smoothing algorithm is used to obtain the directional mean of the uneven color area,and an image with more uniform and full color is obtained.Experimental results show that on standard test images,compared with KSVD,BM3 D,PCA-LPG,EPLL,STROLLR and TWSC algorithms,the proposed SVC-NL algorithm can effectively remove noise and messy textures,the proposed SVC-NL algorithm can achieve better denoising effect on the premise of ensuring image information.In the test images of Dongba painting,the DMS algorithm in this thesis can make the color of Dongba painting more full,bright and uniform,and the denoising effect is better.