Adaptive Correction Method For Image Denoising

Image restoration is an important and challenging topics in image prcessing.Image denoising is an important content of image restoration, Denoising results havea direct impact in the entire image restoration. Therefor image denoising is verysignificance to study. Resently，the most widely used model for image denoising istotal variation model，this model was first proposed by Rudin、Osher and Fatemi. Andit is essentially through Lxto approximate the original problem.The main research in this thesis is the adaptive image denoising in image restora?tion with Zo’us adaptive Lasso ideas to corrected Lx model. In order to correction theprevious model we add a linear term to the total variation Lx model. The role of thelinear term is to turn the vector in the result that who should be zero but just close tozero into zero. So，after correction the final must be more closer to the real value. Inthe specific calculations, we took two stages method: stage one，we solve the wholevariational model，to obtain an approximate solution; Stage two, use the approximatesolution as the initial solution, to solve the corrected model，then get the final result.For the model algorithm，here we use alternating direction method, also get theconvergence of this algorithm. The numerical results show under the same ambiguityand precision, the SNR results has significant improvement. There are five chapters inthis paper:The first chapter introduces the research background、history and progress of theimage denoising,and also brilfy describes the main work of this article.The second chapter describes the total variation model and associated knowledgeand alternating direction method;The third chapter introduces the classical image denoising model，then for thedeficiencies of the Lx model, proposed the adaptive correction TV model;Chapter four presents an alternating direction method for adaptive correction TVmodel，and also proved the convergence of this method.Chapter five gives the relevant numerical experiments. Numerical results showthat the SNR result of the adaptive correction7T model is better than the original one.