基于视点相关性的三维视频中间视点感兴趣区域提取算法.docVIP

  基于视点相关性的三维视频中间视点感兴 趣区域提取算法# 周晓亮，蒋刚毅，邵枫，郁梅，孙越** 5 10 （宁波大学信息科学与工程学院，浙江 宁波 315211） 摘要：给出了一种新颖的感兴趣区域提取结构—类 HBP 结构，并基于这种结构利用相邻视点 的空域相关性和当前视点的深度信息提出了一种三维视频中间视点感兴趣区域快速提取算 法。首先，分别求取左右视点同一时刻彩色图像帧的感兴趣区域，并将左右视点同一时刻彩 色图像帧的感兴趣区域掩膜进行并运算得到的新的感兴趣区域；然后，将中间视点相应图像 帧的原始深度图进行前背景分割；最后，利用逻辑运算获得中间视点对应时刻图像帧的感兴 趣区域。实验结果表明，在保证感兴趣区域提取精度的基础上，所提出算法明显降低了计算 复杂度。 关键词：三维视频；中间视点；感兴趣区域；视点相关性 中图分类号：TN 919. 8 15 Region of Interest Extraction Algorithm of Intermediate View Based on Inter-view Correlation in 3D Video Zhou Xiaoliang, Jiang Gangyi, Shao Feng, Yu Mei, Sun Yue (Faculty of Information Science and Technology, Ningbo University, ZheJiang NingBo 315211) 20 25 30 35 40 Abstract: Based on inter-view correlation in three dimensional (3D) video and a quasi- hierarchical B picture (HBP) structure, a region of interest extraction algorithm of intermediate view is proposed with color and depth information. First, the regions of interest of color videos in left and right views are detected. Then a union operation for the regions of interest of color videos is operated to obtain new regions of interest. Second, the original depth video of the intermediate view is segmented into foreground and background areas. Finally, some logical operations are used for the regions of interest of color video in left and right views and the foreground and background areas of depth video in intermediate view so as to extract interest of the color video in intermediate view. Experimental results show that the proposed method can reduce computational complexity in region of interest extraction of intermediate view, while detecting accuracy is maintained. Keywords: three dimensional video; intermediate view; region of interest (ROI); inter-view correlation 0 引言 为了进一步探索诸如基于内容的视频编解码等交互式多媒体应用技术的潜在能力，运动 图像专家组 MPEG-4 引入了视频对象的概念[1]，并在此方面进行了一些研究。视频对象是 通过视频对象的分割提取获得的, 它不仅是实现基于内容的视频编码的前提条件, 也是实现 基于内容的视频检索和交互式多媒体应用所必需的，更加可以对视觉信息进行主观地描述、 合成、交互[2]。因此，对象提取技术在视频编码、模式识别、视频检索等领域有着越来越 广泛的应用[3]。 近年来，国际上不断有新的对象提取方法提出，Sum 提出一种基于快速 Sna