项目名称: 基于感兴趣区域的互联网级在线视频拷贝检测技术
项目编号: No.61273247
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 张冬明
作者单位: 中国科学院计算技术研究所
项目金额: 80万元
中文摘要: 随着互联网的快速发展,数字视频数据量呈爆炸性增长,特别是互联网视频发展迅猛,对其进行有效检索、管理,以及版权保护已经成为十分紧迫的问题。本项目利用基于视觉特征的统计分析手段,研究流式视频画面的动态聚类算法,以滤除低信息量视频帧和时域冗余信息;研究视频中人所在区域及显著区域两类区域检测算法,以获取视频画面中感兴趣区域。进一步引入区块空间关系验证以研究感兴趣区域内紧致、高效的特征描述算法,并基于GPU研究感兴趣区域提取加速算法、特征提取加速算法,以及空间关系校验加速算法,为面向互联网的在线视频拷贝检测应用提供核心算法和技术。
中文关键词: 显著区域检测;测度排序;空间关系验证;二进制描述子;特征匹配
英文摘要: With the rapid development of Internet,the digital video goes into fast growth period, expecially the web video blurst.It is very urgent to retrieve,manage the videos and protect their copyright.In this project,we will firstly study the dynamic cluster algorithm for the frames of continuous query video stream to remove temporary redundent frames and the ones with very low information. Then to locate the region of interest, we will study two kinds of region detection algorithm based on human detection and salient region detection.Furthermore, spatial relation will be introduced into region feature description and one kind of compact and efficient feature will be used to describe the ROI.To speedup the copy detection, we will also study parallel algorithms for complex modules based on GPU. We hope this project will provide core algorithm for web-scale online video copy detection.
英文关键词: Salient Region Detection;Ordinal Measure;Spatial Verification;Binary Description;Feature Matching