一种基于时间导向的DIBR空洞填补方法  

Hole-filling Method Based on Time-sequence in DIBR

作　　者：丁陈敏 吕昭宏 姚剑敏[1] 许胜允 林博瑛 郭太良[1] 

机构地区：[1]福州大学场致发射国家地方联合工程实验室,福建福州350000 [2]冠捷显示科技(厦门)有限公司厦门创新中心,福建厦门361000

出　　处：《电视技术》2015年第3期68-72,共5页Video Engineering

基　　金：国家"863"计划项目(2012AA03A301;2013AA030601);国家自然科学基金项目(61101169;61106053);福建省自然科学基金项目(2011J01347)

摘　　要：在立体成像系统中,填充新视点合成时图像中产生的空洞是虚拟视点合成技术的重要环节,已有方法是对空洞区域周围的像素进行水平的推测运算来得到填洞像素,其正确率不高而且容易造成图像边缘失真。提出一种基于时间导向的DIBR空洞填补方法,首先通过结构相似帧的方法得到与关键帧拥有相同背景的前后帧图像,然后比较空洞位置对应的前后帧图像的深度值,利用背景信息对空洞区域进行填补。实验结果表明,新方法得到的填补后的图像效果可以较为真实地还原特定的场景,与原始视点的结构相似性可达到91.36%,比传统方法提高了8.38%。In order to fill the holes area in view synthesis, a new hole-filling method based on time sequence in depth image based rendering （DIBR） is developed. In traditional hole-filling method, the hole-filling pixels are calculated from the pixels around the holes. In this study, a new method is developed, the similar images of the holes position are found from the SSIM comparison by time sequence images from the video. And the similar images are selected by SSIM to fill the holes area, the result of the experi- ment shows that this new method can find exact pixels to fill the holes area and improve the image quality and distortion. Finally, the structural similarity index of new method can reach 91.36% and it has 8.38% improving compared with the traditional hole- filling method.

关 键 词：基于深度图像的绘制(DIBR) 空洞填充 结构相似性比较(SSIM) 

分 类 号：TN27[电子电信—物理电子学] TP3[自动化与计算机技术—计算机科学与技术]