Image And Video Inpainting Based On Mixed Criminisi Algorithm

Digital image and video repair is a computer technology that uses effective al-gorithms to automatically repair designated damaged areas in an image or video.Repair technology has a wide range of applications in our daily lives,such as the restoration of books and murals;the restoration of old photos and movies;target removal and restoration of losses caused by compression and transmission.With the improvement of people's living material conditions,people's needs are becoming more and more diverse.Therefore,digital image and video restoration have broad application prospects and become one of the popular research fields in computer vision,image processing and video processing.Through research on restoration algorithms at home and abroad,it is found that Criminisi algorithm can complete large-area damaged image restoration work,but this algorithm still has shortcomings.This article improves the al-gorithm and applies it to image restoration and video restoration.The improvements in this article are as follows:(1)Before repairing,the Krig-ing model is used to pre-fit the damaged area of the image.This processing can fuzzyly restore the original appearance of the damaged area and eliminate the influence of the damaged edge when detecting the edge of the image.(2)As the iteration of the algorithm,the confidence suddenly decreases to close 0,which reduces the performance of the confidence item that measures the reliability of the image block.This paper proposes a confidence improvement function,which uses the difference of confidence under different size neighborhood windows of the same boundary point to measure whether the image block has higher credibility.If the block is more trustworthy,a parameter is introduced to increase the level of confidence.(3)The Criminisi algorithm selects image blocks with a fixed scale as the repair unit.Under different image texture structures,this method cannot adaptively adjust the image block size.In this paper,the facet model is used for edge detection,and four directions of 0 degrees,90 degrees,45 degrees and 135 degrees are selected.The structure of each image block can be measured by the facet model,and the texture image is repaired by selecting larger scale block-s.(4)The matching strategy of Criminisi algorithm to find the optimal matching block is too simple.The matching function only uses the gray scale of the image and does not consider the structural information of the image itself.The optimal matching block found may be inconsistent due to the structural dissimilarity.In this paper,the facet structure information is added to the gray scale measure-ment function SSD,and a WSSD minimization matching strategy is proposed,which emphasizes the importance of image structure while considering the gray scale.The above-mentioned improved algorithm improves the performance of the Criminisi algorithm.In order to make full use of its advantages,the algorithm is applied to video repair.The works done are as follows:(1)Make full use of the time correlation of video.Inter-frame images in the same scene can provide more available information.Therefore,the search range of the optimal matching block is expanded to multi-frame images.(2)This method is suitable for processing videos with slow camera movement or videos with slow target movement.Using the high similarity and slow variability between video frames,the optimal match-ing block of the block to be repaired is searched from the local neighborhood of multiple frames.To ensure sufficient matching data,an adaptive adjustment of the search window scale is proposed.This can not only find effective matching blocks but also effectively compress the time,which is of great significance for repairing patches in old movies.The experimental comparison shows that the image repair based on the im-proved mixed Criminisi algorithm and the fast video repair using the improved mixed Criminisi algorithm are more prominent in the repair effect.