On Image Inpainting Algorithm Based On Sparse Representation

Since Mallat first put forward the theory that image can be sparse representation, sparse representation theory has been widely used in the field of image processing. Image inpainting, is according to the known information to estimate the unknown pixels in the image, making repaired image close to or achieve the visual effect of original image, it is a basic problem in the field of image processing. In recent years, combined sparse theory with image inpainting, has became a research direction in image processing. Therefore, this thesis based on the study of sparse representation theory, united the concept of clustering, researched and proposed some new ideals to solve image inpainting problem.The main research works of this thesis are as follows:1、Introduced the basic concept of sparse representation, discussed the main algorithms to solve the problem of sparse approximation and the main dictionary algorithms in sparse representation.2、Introduced the image inpainting problem based on sparse representation, researched the image inpaingting algorithm based on K-SVD dictionary.3、Based on the image inpaingting algorithm with K-SVD dictionary, proposed an improved image inpainting algorithm based on K-SVD dictionary. The algorithm based on sparse and redundant representations theory, introduced the concept of clustering, considered the similarity between the image blocks. Used the valid information in the damaged image, employed fuzzy C-means(FCM) clustering algorithm classified image blocks, and respectively adopted K-SVD algorithm to get the suitable dictionaries, in order to filled the missing pixels in image blocks, inpainted the damaged image. Experiment results showed, this algorithm compare with K-SVD inpainting algorithm’s results, not only kept the structure information of image, but also better repaired the details of image, especially for the image with more structure information, the restoration results all the better.4、Considered the whole information of image and the similarity between the image blocks, proposed an image inpainting algorithm based on FCM clustering and K-SVD dictionary. This algorithm compensate for the shortage of the quality of image inpainting hardly developed quickly when iteration after reached a certain number of times. Exploited the advantages of dictionary algorithm which training all blocks, and considered the similarity between the image blocks, combined clustering algorithm with sparse theory, realized the image inpainting. This algorithm put the whole dictionary as the base dictionary of training the kinds of image blocks, which came from trained all blocks with K-SVD dictionary algorithm. Then adopted K-SVD dictionary algorithm to train all kinds of image blocks after clustered, got the portion dictionaries which adjusted to each kind of image blocks, calculated sparse representation coefficients, updated all kinds of image blocks, then obtained the final repaired results. Experiment results indicated, the algorithm can better restored the damaged image, particular for the image area with a fine lines, has feasibility and better robustness.