Research On Image Denoising Algorithm Based On Similar Image Patch

Image is the main way for human to transmit visual information and quickly obtain effective information.However,there are many uncontrollable factors in the process of acquisition,transmission and storage of digital image,such as the change of environmental light,the defect of imaging system and the influence of many inevitable human factors,which lead to image quality defect such as noise,fuzzy ghosting,which affects the transmission and expression of real image information.At present,with the rapid development of artificial intelligence,it is necessary to provide better quality image,however sometimes the process of image degradation is inevitable and irreversible,so it is important to restore the damaged image.The traditional image noise reduction and restoration method cannot effectively restore the image texture details,and some methods are too smooth and result in the loss of useful image information,which has certain limitations.In recent years,the denoising methods based on image block have been widely used in various fields of image processing,and it can achieve better results.The existing typical image denoising methods,such as non local mean algorithm,image restoration algorithm based on Gaussian mixture model and so on,are the algorithms with better denoising effect in recent years.However,there are still some defects,such as noise interference to similar information in the image,low similarity of image blocks in some regions,and loss of texture details in the image,resulting in the performance of image denoising and restoration is generally poor Ideal enough.In this thesis,the similarity between image blocks is used,combined with image block clustering and image block reordering algorithms.For the purpose of improving the noise reduction performance of the algorithm,obtaining a high-performance image noise reduction recovery method,and then obtaining high-quality recovered images,has important research significance and application value in practical scenes such as medical treatment,machine vision,and security traffic.The main work of this thesis is summarized as follows:1.Summarize the research progress of image noise reduction related theory and technology,and give a degradation model based on image blocks.In-depth research on the widely used image block-based image noise reduction ideas,combined with image self-similarity features,image block clustering and image block rearrangement,etc.,laid the foundation for the thesis algorithm.2.In view of the classic GMM block prior model that smoothes the image quality after image noise reduction and recovery,existing related algorithms generally have no obvious noise removal effect,unobtrusive image detail information,and complex algorithms when processing images that contain large pollution Defects such as high degree.Drawing on the multivariate Gaussian mixture model,this thesis proposes a method of image denoising based on clustering of image blocks.The algorithm fully considers the self-similarity between the image blocks,and estimates the similarity of the Gaussian clusters formed by the image blocks and their corresponding similar pixel clusters,and calculates the image reconstruction aggregation weight to restore the image contaminated by noise.The experimental results show that the proposed algorithm model has the advantages of better fitting the statistical characteristics of natural images compared with the existing algorithm.Through the objective image quality evaluation of the image denoising results of the algorithm simulation experiment,the effectiveness of the proposed denoising restoration method based on image block is proved.3.In order to solve the problem that the distance of image blocks is not fully utilized in the process of rearrangement,the similarity measure between image blocks in different classification areas is determined by a unified weight,and a NLM image denoising algorithm is proposed based on image block rearrangement.The algorithm takes full account of the self similarity of image blocks in different feature areas in the image.By classifying image blocks into two categories: flat area and edge texture area.The correlation coefficient is introduced into the blocks of edge or detail area to improve the Euclidean distance in NLM algorithm,improve the accuracy of Euclidean distance measurement and obtain the accurate weight suitable for different block areas.The experimental results show that the proposed algorithm has better edge preserving ability and noise reduction effect than the original NLM algorithm.