融合纹理信息的深度图像修复  

Depth maps inpainting with fused texture information

作　　者：王殿伟 陈鹏 李大湘 刘颖 许志杰[3] 王晶[3] WANG Dianwei;CHEN Peng;LI Daxiang;LIU Ying;XU Zhijie;WANG Jing(School of Telecommunication and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;Key Laboratory of Electronic Information Application Technology for SceneInvestigation,Ministry of Public Security,Xi’an,710121,China;School of Computing andEngineering,University of Huddersfield,Huddersfield,UK)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]公安部电子信息现场勘验应用技术重点实验室,陕西西安710121 [3]哈德斯菲尔德大学计算机与工程学院,Huddersfield英国

出　　处：《系统工程与电子技术》2019年第8期1720-1725,共6页Systems Engineering and Electronics

基　　金：国家自然科学基金(41874173,61571361);陕西省自然科学基础研究计划基金(2018JM6118);陕西省国际合作交流项目(2017KW-013);西安邮电大学创新创业项目(2018SC-08);西安邮电大学创新基金(CXJJ2017058)资助课题

摘　　要：针对低质量深度图像中存在的空洞和噪声问题,提出了融合纹理信息的深度图像修复算法。首先利用形态学操作对空洞进行优化,基于区域的分割算法完成空洞区域分割;然后将提取出的空洞区域进行纹理信息填补,分析空洞区域与同场景彩色图像的局部结构相似性,完成空洞初修复;最后利用灰度级图像重建算法,对边缘空洞区域进行填充和平滑处理。基于标准数据集Middlebury,所提算法与快速行进算法、自适应中值滤波算法和形态学重建算法相比,对大面积空洞信息完成了良好的修复。在获得较好修复效果的同时,该算法保持了图像的结构完整性和整体平滑性。To solve the problem of holes and noise in low-quality depth maps,a depth maps inpainting algorithm based on fused texture information is proposed.Firstly,the cavity is optimized by morphological operation,and the segmentation of the cavity is completed by a region-based segmentation algorithm.Secondly,the extracted cavity region is filled with texture information,and the local structural similarity between the cavity region and the same scene color image is analyzed.Finally,the gray scale image reconstruction algorithm is utilized to fill the edge hole region and smooth the region.Compared with the fast marching algorithm,the adaptive median filtering algorithm and the morphological reconstruction algorithm on the standard dataset Middlebury,the proposed algorithm completes the large-area cavity information.While getting better inpainting results,the algorithm maintains the structural integrity and overall smoothness of the image.

关 键 词：深度图像修复 空洞填充 形态学处理 区域分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]