基于结构-纹理分解的数字图像分层修复算法  

Digital Image Layered Restoration Algorithm Based onStructure-Texture Decomposition

作　　者：孙宇 李晗[2] SUN Yu;LI Han(School of Information Engineering,Xinjiang Institute of Technology,Aksu Xinjiang 843000,China;School of Electronic and Information Engineering,Liaoning University of Technology,Jinzhou Liaoning 121000,China)

机构地区：[1]新疆理工学院信息工程学院,新疆阿克苏843000 [2]辽宁工业大学电子与信息工程学院,辽宁锦州121000

出　　处：《计算机仿真》2023年第6期220-224,共5页Computer Simulation

基　　金：辽宁省博士科研启动基金计划项目(2019-BS-121);中央引导地方科技发展专项资金(2020JH6/10500067)。

摘　　要：采用目前方法进行数字图像修复时,没有对边缘的衔接效果和灰度跳变问题进行考虑,导致视觉效果差、SSIM与PSNR值低的问题。提出基于结构-纹理分解的数字图像分层修复算法,首先运用TV-L~1模型完成数字图像的结构-纹理拆分,然后利用快速行进法对图像进行预填充,最后对数字图像结构-纹理部分进行分层修复,在结构部分运用结合12个邻域点的加权组合更新受损点,获取更好的衔接效果,纹理部分采用Criminisi算法,用局部搜索代替全局搜索,完成数字图像的分层修复。通过实验结果可以看出,所提方法能够有效地改善视觉效果,提高SSIM与PSNR值。When the current method is used for digital inpainting,the problem of edge connection effect and gray scale jump is not considered,resulting in poor visual effect and low SSIM and PSNR values.Therefore,a layered res-toration algorithm of digital image based on structure-texture decomposition was proposed.First of all,TV-L'model was used to decompose the structure and texture of digital image.And then the image was pre-filled based on the fast-marching method.Finally,the structure and texture of digital image were repaired layer by layer.In the structure part,the weighted combination including twelve neighborhood points was used to update the damaged points and to ob-tain better connection.In the texture part,Criminisi algorithm was used to replace the global search with local search.Thus,the layered restoration of digital image was completed.Experimental results show that the proposed method can effectively improve the visual effect and increase SSIM and PSNR values.

关 键 词：结构-纹理分解 数字图像 分层修复 快速行进法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]