Digital Image Denoising Algorithms Based On Patch

Image is an important means to obtain information,but the image will be mixed noise during collection and transmission process,which seriously interferes with the image information and application value.Therefore,image denoising has important research significance.The image is divided into patches so that the image can be refined to get a better denoising effect.Based on the image patches,this thesis studies the similarity and sparseness of image to denoise.The main research work in this thesis is as follows:(1)Fast Patch Group Based Nonlocal Self-Similarity Prior Learning for Image Denoising denoising algorithm(PGPD)research.PGPD algorithm has a good denoising effect,and this thesis proposes a fast PGPD denoising algorithm to improve denoising speed.In the process of searching non-local similar patches,the size of the search window is adjusted by non-local self-similarity of the patches,and the non-local similar patch range is reduced by sampling the search patches.Then the non-local similar patches corresponding to the target local patch can be defined efficiently and accurately by doing euclidean distance calculation,which reduces the difficulty of calculation.Experimental results of digital images show that compared with traditional PGPD algorithm,the denoising speed of fast PGPD algorithm can be greatly improved with a little loss of image.(2)Sequential Generalization of K-means image denoising algorithm(SGK)combined with principal component analysis of noise estimation.SGK algorithm has the characteristics of fast denoising speed and excellent denoising performance.However,the noise standard deviation must be known in advance when using SGK algorithm to denoise image.This thesis presents a denoising algorithm combined with SGK dictionary learning and the principal component analysis noise estimation.At first,the noise standard deviation of the image is estimated by PCA noise estimation algorithm.And then the estimation is used for SGK dictionary learning algorithm.According to the denoising experiments of multiple images with different noise standard deviation,the results show that SGK algorithm has the best denoising effect compared with the other three dictionary learning algorithms,and the SGK combined with PCA is superior to the SGK combined with other noise estimation algorithms.Compared with original SGK algorithm,the proposed algorithm has higher PSNR and better denoising performance.Therefore,this algorithm is feasible for image denoising with unknown noise standard deviation.(3)External target database image denoising(TID)based on block-matching and 3Dfiltering algorithm.TID algorithm denoise the image by external target database.Compared with the traditional image denoising algorithms,the TID algorithm has better denoising performance.However,the TID algorithm repeatedly traverses the external target database during the denoising process,which has certain blindness and takes a long time.So this thesis proposes a database image denoising algorithm based on BM3 D,which is TID-BM3 D algorithm.In the initial denoising stage,the algorithm uses TID to externally denoise the original image information.Firstly,it searches similar image patches in the target database to fully supplement the similarity information of the image patches,and then it determines the estimated value through group sparsity and local priors.In the second denoising stage,BM3 D is used for denoising.Patch-matched collaborative filtering is used to quickly search for similar patches,and the original image information is estimated accurately in a short time.The proposed TID-BM3 D algorithm is applied to text database denoising and face database denoising.The experiment results show that the proposed algorithm has better denoising performance than the original TID algorithm,and it has been improved both in denoising speed and denoising performance.