优化改进的深度图像空洞填补算法  

Improved depth image hole filling algorithm

作　　者：高雅平[1] 陈一民[1] 邹一波[1] 黄晨[1] 高明柯[1] 李启明[1] GAO Ya-ping CHEN Yi-min ZOU Yi-bo HUANG Chen GAO Ming-ke LI Qi-ming(School of Computer Engineering and Science, Shanghai University, Shanghai 200444, Chin)

机构地区：[1]上海大学计算机工程与科学学院,上海200444

出　　处：《计算机工程与设计》2016年第11期3040-3044,共5页Computer Engineering and Design

基　　金：上海市科委基金项目(12510708400);上海市自然科学基金项目(14ZR1419700);2015年度上海大学电影学高峰学科基金项目(n.13-a303-15-w23)

摘　　要：鉴于Kinect获取的深度图像边缘易出现空洞的问题,提出一种优化改进的深度图像空洞填补算法。通过引入与深度图像同步的RGB图像,使两者配准对齐;基于RGB图像搜索空洞候选最优匹配点,通过RGB图像边缘轮廓将空洞区域进行划分以限定最优匹配区域,采用反映射方式获得深度图上的空洞最优匹配点。实验结果表明,相比原有的Criminisi算法,包括改进的Criminisi算法,本文算法能更准确地填补深度图的空洞,同时保持修复区域边缘部分的平滑性。Image taken using Kinect tends to have holes.An improved hole filling algorithm was proposed to solve this problem.A synchronized RGB image was introduced,which was registered with the depth image.The candidate optimal matching hole points were searched based on RGB image,the region of hole in depth image was divided by extracting the edge of RGB image to restrict the best matching hole-region,the optimal matching hole points were got in the way of reversible map.Experimental results demonstrate that,compared with the traditional Criminisi algorithm including the optimized Criminisi algorithm,the proposed algorithm can fill the hole more accurately while maintaining the smoothness of fixed edge region.

关 键 词：Crimininsi 深度图 空洞 RGB-D图像配准 最优匹配 边缘轮廓 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]