Research Of Image Completion Based On Exemplar Patch

Digital image completion or image inpainting means that a computer software or program can reconstruct the missing part of an image according to incomplete known information in the image,so that the resulting image has good visual rationality.This technology has high application value in the aspects of heritage restoration,super resolution reconstruction,image compression and coding,and the film post-production.With the development of electronic technology and the popularity of portable camera equipment in recent years,the demand for the accuracy and real-time of image completion and image editing is growing.This thesis focuses on exemplar-based image completion.As existing novle completion methods often have the problem that texture and structure is disordered,and it always takes a long time when the image missing part is large.In order to improve the quality of image and reduce the completion time,in this thesis,new priority assignment and exemplar patch searching range strategy are proposed to make the image completion effect better.The main contribution and innovation of this thesis are as follows:(1)An image completion algorithm based on redefined priority and image division is proposed.The exemplar-based image inpainting method Criminisi algorithm has poor inpainting effect sometimes since the confidence term declines to zero easily which makes the priority invalid and disorders the inpainting sequence.Firstly,the confidence term in the priority is redefined,the chessboard distance in the exemplar patch is used to replaces the original calculation formula,so that validating the priority and reducing the matching error caused by unreasonable inpainting order.Otherwise,the excessive scope,low efficiency and non-visual texture matching problem exist in searching matching patches,the image is divided into different size blocks according to the image texture information,so that the exemplar patches to be completed are only searching in the image block region that has similar colors and textures.The proposed algorithm has good performance in reducing the mismatch and the accumulated error in the completion algorithm,and the time consuming is reduced.(2)An image completion algorithm based on saliency map and gray entropy is proposed.In the first phase of the completion algorithm "Exemplar Based Image Inpainting Using a Modified Priority Definition",the completion sequence is determined by structural information,which is easily disturbed by the image texture information and is difficult to find the candidate patches with high priority.In this part,the human visual attention information saliency map is added to the stage of priority,which ensure that the parts with strong structural information and visual importance are completed preferentially,so as to ensure the coherence of the edge structure in the image.Secondly,according to the feature that the saliency of the patch to be completed should be consistent with source patch,the saliency map information is added to the matching process of the sample to make it match with the color information to improve the matching precision.Finally,aiming at the problem of insufficient sample source caused by the fixed neighborhood of the patch to be completed as the search area,we use gray entropy to dynamically adjust the size of search areas,not only effectively solve the problem of the number of samples,but also improves the search speed.The proposed algorithm is better than the original algorithm in preserving the structural integrity and completion effect.