Research On Image Inpainting Algorithm Based On Sparse Representation

With the widespread use of computer and multimedia technology,image-based information becomes the mainstream media of information exchange,which greatly affects the lifestyle of people.As an important branch of image processing,image inpainting has been widely used in medicine,military and public administration.In recent years,image processing technology based on sparse representation has been widespread concerned.This thesis is based on the theory of sparse representation to sparse the image signal firstly,and then reconstruct the original image using a reconstruction algorithm.The study of this paper mainly focuses on the selection of dictionaries in the process of image inpainting and a solution with morphological component analysis decomposition.The main work includes the following aspects:(1)The images consist of complex two-dimensional signals,which can be transformes into a linear combination of a bit of atoms.This paper introduces some kinds of traditional models for image inpainting and sparse representation image inpainting algorithm.In order to overcome the shortcomings such as slow speed of image restoration and long iterative time,we choose the K-SVD dictionary as the basic of the image inpainting.In this paper,we firstly divided the image into some patches,using the information of effective areas in the images.And then,we use the method of Steering Kernal Regression Weight to cluster these patches,giving the most suitable K-SVD dictionary to each portion,repairing the damaged images.Experimental results show that our algorithm can repair the damaged images better than traditional algorithms,such as the images with noisy and so on.(2)Aiming at the problem that the image structure and texture components can not be ued well in the current image inpainting,this paper studies the modle of morphological component.In order to show the information of the two parts,this paper uses redundancy discrete wavelet transform dictionary and the concatenation dictionary of wave atomic dictionary and discrete cavelet transform dictionary to replace the original algorithm.At the same time,accroding to different characteristics between the cartoon part and the texture part,this paper using the total variation method and the Bayesian weight method to restore them separately.Finally,this paper adds the restored structure information and texture information together,and the finalrestoration image can be obtained.The experiments show that our methods can get better restoration results from the damaged images.