Research On Image Denoising Algorithms Based On Dictionary Optimization Strategy

Noise pollution of images is usually generated during the acquisition and transmission of image data.Noise in images often affects subsequent image processing,including image coding,feature extraction,and target detection.Image noise reduction is a key process for other image processing tasks,which can effectively improve the processing effect of subsequent image processing techniques.Image denoising algorithms based on sparse representation and dictionary learning have become a part of the most important and most advanced denoising algorithms.The thesis describes and improves the existing algorithms based on sparse representation and dictionary learning.The specific research contents of thesis are as follows:1.The classical K-singular value decomposition(K-SVD)algorithm uses a dictionary to sparsely represent images,which can denoise an image as well as maintain the effective information of the original image.However,the algorithm performs poorly under strong noise,so an image denoising algorithm based on dictionary update and dictionary atomic optimization is proposed.Firstly,the weighted sequential dictionary learning(SDL)method is used to replace the K-SVD algorithm,which can obtain a more sparse representation of the image dictionary.Secondly,we use the image structure characteristics of the dictionary atom after training,and fully consider the influence of the structure complexity of the original image and noise intensity on the dictionary atom to adaptively detect atoms and deletet noise atoms among them.Finally,the reconstructed image is reconstructed using the optimized dictionary.Experimental results show that the proposed algorithm can achieve better denoising effect than the classical K-SVD denoising algorithms.2.The traditional sparse representation algorithm achieves the purpose of denoising by externally a priori global training dictionary,or by performing dictionary learning by a noisy image itself.However,neither of them can represent an image effectively.Therefore,this thesis proposes an image denoising algorithm based on adaptive learning dictionary and global training dictionary.Firstly,the K-SVD algorithm is used to learn the dictionary.Secondly,the relevant atom detection will be performed on the dictionary,and choose the dictionary atoms in the dictionary which can not represent image reasonably,and then,the corresponding image blocks in the noisy image will be found through these dictionary atoms,and they will be dealt with the global training dictionary.Finally,the restoration image is reconstructed by these image blocks.Experimental results show that the proposed algorithm combines the advantages of both dictionary processing methods in some extend,and achieves better denoising effect compared to the original algorithm or other algorithms.