纹理萃取耦合U型更新的图像修复算法  

Image Inpainting Algorithm Based on Texture Extraction Coupled with U-shape Updating

作　　者：唐宝燕[1] TANG Bao-yan(School of Information Engineering,Weifang Engineering Vocational College,Qingzhou 262500,China)

机构地区：[1]潍坊工程职业学院信息工程系,山东青州262500

出　　处：《辽东学院学报（自然科学版）》2020年第3期205-210,共6页Journal of Eastern Liaoning University:Natural Science Edition

基　　金：2019山东省潍坊市科学技术发展计划项目(2019GX098)。

摘　　要：针对传统图像修复算法在对破损区域优先权计算时,没有对破损图像纹理进行合理的预先处理,致使修复后的图片出现错误的匹配和纹理延伸等诸多问题,提出一种纹理萃取耦合U型更新的图像修复算法。首先,利用小波变换的方法去噪再结合角点检测分析,萃取出破损区域的纹理指标,对破损区域进行纹理萃取处理,以有效反应破损区域图像与匹配图像中纹理的相似度,在优先权计算时,利用萃取的纹理指标,以增加对匹配块优先选取的可靠性。其次,为解决Criminisi算法中置信度值快速趋0,从而导致修复图像连贯性差的问题,利用U型计算方式对修复块的置信度进行强化从而抑制置信度的快速衰减,解决修复图像连贯性差的问题,优化图像修复的视觉效果。通过与不同的图像修复算法进行对比实验,验证了文中所提修复方法具有更加稳定修复效果,更加符合肉眼在对图像观察时的视觉要求。In order to solve the problem that the texture of the damaged image is not pre-processed reasonably and effectively when the priority of the damaged area is calculated by the traditional image restoration algorithm,which often leads to the wrong matching and texture extension of the repaired image,this paper proposes an image restoration algorithm based on texture extraction coupled with U-shape update.In this method,firstly,the damaged area is processed by texture extraction,and then the texture indexes of the damaged area are extracted by wavelet transform and corner detection analysis.These indexes can effectively reflect the similarity between the damaged area image and the matching image,and the reliability of the priority selection of the matching block can be increased by combining the extracted texture indexes in priority calculation.Then,in order to solve the problem that the confidence value in the algorithm tends to zero rapidly because of continuous iteration,which leads to the poor consistency of the repaired image,this paper uses the U-type calculation method to strengthen the confidence of the repaired block so as to suppress the rapid decay of the confidence value,solve the problem of poor consistency of the repaired image,and optimize the visual effect of image restoration.By comparing with different image restoration algorithms,it is verified that the proposed method has a more stable restoration effect,and more in line with the visual requirements of the naked eye when observing the image.

关 键 词：图像修复 Criminisi算法 置信度 优先权计算 连贯性 视觉效果 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]