面向编码和绘制的多视点图像深度估计  

Multi-view image depth estimation for coding and view rendering synthesis

作　　者：张秋闻[1] 安平[1,2] 张艳[1,3] 张倩[1] 张兆杨[1,2] 

机构地区：[1]上海大学通信与信息工程学院,上海200072 [2]上海大学新型显示技术及应用集成教育部重点实验室,上海200072 [3]安徽财经大学计算机科学与技术系,安徽蚌埠230030

出　　处：《光电子．激光》2011年第10期1569-1574,共6页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金重点资助项目(60832003);国家自然科学基金资助项目(60902085;61172096;61171084);上海市教委科研创新重点资助项目(09ZZ90);上海市科委重点项目基金资助项目(10510500500)

摘　　要：针对自由视点三维电视系统中深度估计不准确将给后续多视点深度编码和虚拟视点绘制带来困难的问题,提出一种面向编码和绘制的多视点图像深度估计算法。首先对初始深度进行一致性检查,并采用自适应匹配误差策略删除不可靠的匹配以减小初始深度的误匹配。然后根据融合准则将多幅参考深度图合成为一幅深度图以提高深度图的精度。最后,采用多边滤波器在保护深度图细节的同时对影响虚拟视点绘制的对象边界以及低纹理区深度不连续处错误进行矫正优化。实验结果表明,该算法能显著提高深度图的编码效率和减少视点合成的赝像。To measure the influence of depth estimation error on depth coding and virtual view rendering in a free-viewpoint television （FTV） system, a multi-view depth estimation is proposed to solve the problem for coding and view synthesis. Firstly,the consistency of initial depth is checked and the influ- ence of initial miss-matches is minimized by introducing an additional adaptive matching error that penali- zes the unreliable matches. Then according to specific criteria, the multi-reference depth maps are merged into one disparity map to improve the quality of disparity map. Finally,a multi-lateral filtering is used to preserve details in the depth map and simultaneously smooth the depths in occluded areas at object boundary,less texture and discontinuity regions. Experimental results show a significant improvement of the initial input depth maps and coding efficiency,as well as a reduction of view synthesis artifacts.

关 键 词：自由视点电视 虚拟视点绘制 深度编码 多视点深度估计 

分 类 号：TN919.8[电子电信—通信与信息系统]