Research On Image Denoising Based On Sparse Representation And Dictionary Learning

The quality of image denoising directly affects the effectiveness and reliability of the follow-up work of image processing.Therefore,image denoising has become an important part of image processing.Acquiring adaptive learning dictionaries has received extensive attention in the image processing field.Existing image denoising algorithms have the following problems: geometric image edge texture and detail feature information loss,and lack of systematic methods for adjusting parameters,etc.Therefore designing self-adaptive training and dictionary learning image denoising algorithms has important theoretical research significance and practical application value.This thesis aims to tackle the difficulties in image denoising and dictionary learning,it mainly studies issues including geometric texture features loss and the non-global optimization of various parameters in image denoising algorithms.Two kinds of image denoising algorithms based on dictionary learning have been designed,which demonstrate improved performance over state-of-the-art.The main contributions f his paper are as follows:(1)For geometrical edge texture and information loss issue,presented in existing image denoising algorithms.we design an image denosing algorithm by combining the dictionary learning and Bandelet transform algorithms.The Second Generation Bandelet Transform has the regularity of geometric flow and can choose the optimum geometric direction adaptively.It splits with quadtree and calculates its optimal geometry flow to obtain the image block Bandelet.Then,it learns the dictionary the K-SVD algorithm to get the dictionary corresponding to each image block.This procedure repeats iteratively,until the most sparse overcomplete dictionary is achieved.The algorithm can preserve the details such as the sharpness of the image and the geometric edge texture.(2)The above image denoising algorithm used various parameters set by experiences,instead of the global optimal solution of the objective function,therefore lacks a systematic parametre tuning theoretical method.Therefore,an adaptive dictionary learning image denoising algorithm based on Stein Unbiased Risk Estimation(SURE)is proposed.We use SURE to automate the systematic adjustments of parameters,in order to obtain the global optimal sparse solution of the objective function.The dictionary is trained with K-SVD to obtain a dictionary corresponding to the adaptive sparse solution for each image block.For smooth images,a large number of false edges can be removed,and the connectivity of the image are enhanced.