压缩域中叠化镜头的快速检测算法  

A Fast Algorithm to Detect Dissolve Shot in Compression Domain

作　　者：肖友能[1] 薛向阳[2] 

机构地区：[1]复旦大学计算机科学与工程系,上海200433 [2]复旦大学智能信息处理开放实验室,上海200433

出　　处：《计算机研究与发展》2004年第11期1982-1989,共8页Journal of Computer Research and Development

基　　金：国家自然科学基金项目 (60 0 0 3 0 17;60 3 73 0 2 0 ;6993 5 0 10 ) ;国家"八六三"高技术研究发展计划基金项目 (2 0 0 1AA114 12 0 ;2 0 0 2AA10 3 0 65 ) ;上海市政府科技发展基金项目 (0 3DZ15 0 19;0 3DZ14 0 15 )

摘　　要：随着网络和多媒体技术的不断发展 ,基于内容的多媒体信息检索技术变得越来越重要 同成熟的文本检索技术相比 ,视频检索还处在研究和探索阶段 视频检索的一个有效方法是将无结构的视频节目进行镜头分割 ,根据每个镜头的关键帧对视频建立索引 因此 ,镜头分割是基于内容的视频检索的基本步骤 ,在各种类型的镜头检测算法中 ,叠化镜头是很难检测的 根据叠化 (dissolve)镜头内部预测帧预测误差能量和运动矢量分布特点 ,提出一种在压缩域中分割叠化镜头的新算法 与公开发表的同类算法相比 ,它具有以下优点 :工作在压缩域上、速度快、鲁棒性好。With the development of networks and multimedia technologies, content-based multimedia information retrieval is more and more important. Compared with the matured text retrieval technology, video retrieval is still at the initial stage. An effective method for video retrieval is to detect boundaries of shot, then to build index for key frames of each shot. In such case, segmentation is the fundamental process of content-based video retrieval. Among many different kinds of scene change, dissolve is the most difficult for detection. An effective and fast algorithm is proposed to detect dissolve shots in compressed video stream. Firstly it makes use of the distribution of prediction DCT error energy to find candidate dissolve shots. Secondly it adopts motion vectors and direct current (DC) component's histograms to identify those candidates. Experimental results show that the proposed method is much faster, more robust and of higher accuracy in comparison with other methods published.

关 键 词：叠化 镜头分割 信息检索 视频数据库 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]