Research On Digital Image Inpainting Technology

Digital image inpainting is a hot research direction in the field of digital image processing in recent years.The purpose of image inpainting is estimate the missing information in the destroyed area according with designed rules. The algorithm should restore the image as much as possible, this problem's essence is reconstruction of complete information through the known incomplete information. Although image inpainting technique has remained in the semi-automatic step, it has been successfully applied to some multimedia applications and signal processing field, therefore, image inpainting technique has a broad application prospects. Considering the influence of academic, digital image inpainting research's essence is a posteriori decision problems by priori knowledge, effectively estimate the posteriori signal in the situation of limited prior knowledge, this problem has important theoretical significance in the field of artificial intelligence.This paper expand the research focuses on the image inpainting technique based on sample matching and the image inpainting technique based on sparse representation. First of all, the article summarize the basic theories of the two techniques. Then the paper given some new solutions on the basis of these classical algorithm, aiming at the shortcomings of these algorithms, considering two ideas about the optimization of sample matching strategy and the optimization global modeling. Finally, through the analysis of the experiment verify the performance of the new algorithm. The work of this paper is reflected in the following points:1)Firstly, the paper discusses the image inpainting theory based on texture synthesis and the image inpainting theory based on sparse representation, including the basic image inpainting algorithms based on these theories, and analyzed the development of the study of the two inpainting technique at home and abroad.2) Aiming at the disadvantage of low efficiency of the existing inpainting algorithm, the paper design the image inpainting algorithm based on local entropy of image and synthesis sample block. The new algorithm use the image local entropy feature partition image, according to the block to be repair the local feature values to determine the iterative generation of sample space before matching sample, this method can reduce the average search length. At the same time, the algorithm uses weighted synthetic matching block method to obtain the final filling information. The method improves the repair quality of the algorithm, and in the same time it improves the efficiency of the algorithm.3) On the basis of the algorithm in chapter three, aiming at the disadvantage of Euclidean distance in the function of measure similar of complex texture image block, design a new inpainting adaptive algorithm based on multiple matching criterions. First increase a matching criterion using the Bhattacharyya coefficient auxiliary measure, and according to the edge complex of the repairing block, using corresponding matching criterions similarity. At the same time, the algorithm adopts dynamic contrast search algorithm to optimize the run time of the algorithm, reduce the calculation times of similarity criteria in each iteration, which can effectively reduce the time and cost of repair.4) In the basis of global dictionary learning inpainting algorithm, discusses the effects of differences exist between training samples on the repair effect. Paper put forward the idea of classification learning dictionary, first training sample clustering, then, for each class of samples to separate the dictionary learning and complete the repair. The experiment proved the effectiveness of the algorithm.