Research On Improved Exemplar-based Image Inpainting Algorithm

With the rapid development and progress of science and technology,image inpainting technology has become one of the most popular technologies in digital image processing.The purpose of image inpainting is to predict the content of the area to be repaired according to the known content of the complete area in the image,to make the predicted content consistent with the real content as much as possible,and finally to make the inpainted image present a satisfactory visual effect.Image inpainting methods are mainly divided into two categories.One is the image inpainting method based on Partial Differential Equation(PDE).Its main idea is to smoothly spread information from outside to inside along the isophotes direction in the damaged area.It mainly is used to restore some small scratches and spots in images.The other is exemplar-based image inpainting method,which synthesizes textures in a copy-paste manner in image.This method is suitable for filling in the large damaged area.This paper mainly researches on how to accurately search for match patches and how to make full use of the texture and structure information of known regions in the example-based image inpainting method.The experimental results prove the effectiveness of the proposed methods.The main work and innovations of this paper are as follows:(1)This paper proposes an exemplar-based image inpainting method using angle-aware patch matching.Firstly,it uses Moving Least Squares method to fit surface of the damaged area in the 3D subspace to obtain the initial pixel values;Then we use the initialized pixel values of the damaged area as prior knowledge to improve the priority function,and the inpainted order becomes more reasonable;Secondly,a patch matching strategy based on the angle-aware is used to find the best-matched patch for the target patch to be inpainted.This matching strategy not only considers the color characteristics of the exemplar patch,but also introduces gradient information to improve the matching accuracy between two patches.Finally,the missing regions are restored with the most similar patch.The algorithm can restore not only small-scale damaged areas,but also large-scale ones.(2)This paper proposes an image inapinting algorithm based on the improved priority and adaptive patch search.We have improved the priority of the original Criminisi algorithm.Firstly,by introducing gray features,curvature terms,and variances into the data term,the regions with rich structural information are filled preferentially,so that the inpainted order is performed in the correct direction;Secondly,the gradient feature and the Bhattacharyya distance are used to redefine the patch matching criteria.The color distance is also taken into consideration,and the spatial features are also considered.The Bhattacharyya distance is used to improve the similarity between the target patch and the match patch,and also improve the matching accuracy.Finally,a search strategy based on gray entropy and neighborhood gradient information is used to adaptively perform global or local search in the image.Based on sufficient theoretical analysis,this paper conducts a large number of simulation experiments on the proposed methods,and evaluates the image inpainting quality from subjective and objective perspectives,thus verifying the effectiveness and feasibility of the proposed methods.