Research On Image Inpainting Algorithm Based On Sample Block

Digital image inpainting,which is a research hotspot of computer vision and computer graphic,is a technology that fills the known information into the unknown domains of a damaged image.This technology has been applied in digital restoration of ancient paintings for conservation purposes,restoration of old photograghs,video making,photography,transmission for network data and so on.Due to the wide application of digital inpainting techniques,the research of this technique has important theoretical significance and practical value.Based on the research of Criminisi algorithm,this paper proposes an improved algorithm for its shortcomings that uses the confidence item threshold and logarithmic smoothing method when calculating the priority,and uses the Jaccard distance to increase the matching criterion constraint,and adaptively selects the template size according to the image structure information,and changes the search strategy by introduction the method of image segmentation.A more reasonable filling order and block matching effect of the principle are obtained through the above improvement algorithm,and the calculation time is reduced,the main search work of this paper is as follows:(1)Considering the defect that the confidence item drops sharply and tends to zero,the confidence item is improved to logarithmic form,which slows down the decreasing trend of confidence.In order to make the improved confidence level consistent with the original confidence level,the confidence item is increased by 1.At the same time,a threshold demarcation method is proposed.When the confidence item is greater than the threshold value,the logarithmic confidence term is used;when it is less than the threshold value,it's set as a constant,and the priority function is dominated by the data item.In view of the single structure of the data item,the structure tensor is introduced to interact with the data item,and the corresponding coefficient is allocated to repair the structure part first.(2)Considering that the SSD matching principle is too single,the Jaccard distance is introduced to measure the similarity between the matching block and the sample block,and the constraint of the matching criterion is added.Because the new Jaccard metric plays different roles,including fine-tuning and amplification.there are two forms of new matching criteria,and they are compared and analyzed through experiments.(3)The method of fixed sample block size of 9*9 adapted by Criminisi algorithm does not consider the image characteristics of the image block to be repaired,so the method of fixed sample block size is defective.Therefore,this paper considers the image features of the block to be repaired,and adaptively selects the appropriate template size according to the size comparison of the characteristic values of the block image to be repaired.At the same time,a search strategy based on texture segmentation is proposed,and the image is segmented and divided into different image regions,and then the appropriate search area for the best matching block is selected according to the similarity metric L.To sum up,the improved algorithm of priority function,matching criterion,sample block size and search strategy improves the repair effect compared with the original Criminsi algorithm.At the same time,different algorithms are compared through experiments and analysis.In the case of comprehensive repair effect and repair efficiency,the algorithm in this paper is relatively better.