The Research Of Doubly Image Denoising Algorithm Based On K-SVD

The related research of digital image denoising has always been an hot point in the digital image processing erea.However,as a new theory in the digital image processing erea,the sparse signal expansion theories has got great attention from reasearchers because of its own good character.It shows that this method's expression can be easily saved and well pick the original images' substantive characteristics after the sparse representation.Now,this algorithm has already been used in every erea of digital image processing,such as the research of digital image denoising,the research of digital image classification,the fusion technique of digital images,the face recognition technique of digital image etc.This thesis firstly study the Kernel Singular Value Decomposition(K-SVD).This algorithm has a good character of image denoising and adaptivity,however,there always exists the disadvantage of edge blur after denoising the images,what's more,the effect of denoising is not perfect when adding too much noise;the running speed is too slow because of high occupied memory when it works.Based on the former deficiency and the full research of many image denoising algorithm,this thesis comes up with a double image denoising algorithm based on the improved K-SVD algorithm:1、Aimed at improving the long running time and edge blur issues of K-SVD algorithm when adding too much noising,this thesis uses the modified APN algorithm to improve the dictionary made up of by the K-SVD algorithm which is a better algorithm based on the original AP clustering algorithm.What's more,compared with the traditional K-SVD denoising algorithm,we can get a conclusion that the improved image denoising algorithm has distinct advantages on the running speed and denoising effect after simulated analysis.2、The imporved K-SVD algorithm has distinct advantages compared with the original algorithm,however,the denoising effect is not perfect and there always exist blurring phenomena on the edes after restoring the images.To solve the problem,this thesis comes up with a double-image denoising algorithm: Firstly,we use the imporved K-SVD algorithm to denoise the pictures;secondly,use the geodesic distance denoising algorithm to denoise the images again.The original images' noise has mostly been denoised,so the iterations and complexity of the algorithm improve a lot when denoising again.The results of the simulated analysis prove that the double denoising method has better effect on the images compared with the original K-SVD algorithm and geodesic distance denoising algorithm,and the more noise the images have,the better effect could be seen,at last,the PSNR get 8-15 dB more than the former two denoising algorithm.