Research On Exemplar-based Image Inpainting Algorithm

Image inpainting is to repair the missing area existing in the image or to remove the unwanted target object in the image,which makes the inpainted result maintain the visual rationality of the image,that is to say,the missing and inpainting of image should not be observed.With the development of digital technology,image inpainting no longer needs to be manually performed but can be automatically performed by a computer program.Image inpainting is gradually becoming more efficient and automated so as to save time,manpower and material resources.As a research hotspot in the field of computer image processing,digital image inpainting technology has been widely used in the repair of text or scratches,the removal of objects,the production of film and television animation effects,image compression and so on.The current image inpainting algorithms are mainly composed of two representative automatic inpainting algorithms,which are structure-based image inpainting and exemplar-based image inpainting.Through summarizing and analyzing the research theories of image inpainting algorithms in domestic and foreign,this paper proposes an improved algorithm based on the exemplar-based inpainting algorithm.Based on the Criminisi algorithm,the method proposes a reasonable priority calculation formula,a matching way by combining the SSD and physical distance,and a logarithmic confidence update criterion.Experimental results have been indicated that the proposed image inpainting algorithm not only has good inpainting effect but also has high stability.The major research contents of the paper are given as follows:(1)In the priority calculation part,this paper proposes a method to calculate the average priority of the blocks to be matched.Through analyzing the definition of priority in the Criminisi algorithm,it can be found that the role of the data item becomes smaller and smaller along with the iteration progressed,which will result in the error of the inpainting order.So the addition is used to instead the multiplication in the calculation of the average priority for the block to be inpainted.According to the calculated priority value,the matching block corresponding to the maximum value is selected as the priority repair area,and the inpainting order is determined.Experiments show that the inpainting sequence determined in our method can reduce the blurring effect occuring in Criminisi algorithm and improve the accuracy of the inpainting sequence.(2)In the CPT algorithm,the value of the highest priority is used to update the confidence term.However,along with the iterative process progresses,the confidence term's value will drop rapidly and approach zero,which may lead to the wrong inpainting direction.Therefore,the data term has no effect on the calculation of priority so that its inpainting order is not reasonable.To solve this problem,this paper proposed an update method by utilizing the logarithmic update function.Experimental results show that the proposed method reduces the speed of confidence degradation,improves the reliability of the algorithm repair sequence,and obtains reasonable inpainting results.Besides,the algorithm show more efficient and stable.(3)Based on the matching criterion of the Criminisi algorithm,this paper proposes a matching criterion by combining physical distance and SSD.Since the similarity of image blocks in an image includes not only color features but also other features such as physical distance,this paper combines image color features and physical features,and then uses them to optimize the matching criteria.Experimental results demonstrate that the matching accuracy of image blocks improves and the reliability of inpainting consequence increases.