视频层次结构挖掘  

Video Hierarchical Structure Mining

作　　者：付畅俭[1] 李国辉[1] 胡军涛[1] 

机构地区：[1]国防科技大学信息系统与管理学院系统工程系

出　　处：《计算机工程与应用》2006年第26期159-162,共4页Computer Engineering and Applications

基　　金：国家自然科学基金资助项目(编号:60273066)

摘　　要：视频处理的关键是视频信息的结构化,视频基本结构是由帧、镜头、场景和视频节目构成的层次结构。视频层次结构挖掘的一个简单框架是对视频进行镜头分割、抽取镜头特征和视频场景构造。论文在镜头分割的基础上提出了基于多特征的镜头聚类分析和基于镜头的场景边界检测两种视频场景构造方法,从而实现视频层次结构挖掘。实验表明,基于镜头的场景边界检测性能优于基于多特征的镜头聚类分析。To structuralize video streams plays an important role in the processing of video.The basic structure for video is a hierarchical structure which consists of four kinds of components,namely frame,shot,scene,and video program.A simple framework for video hierarchical structure mining is to partition continuous video frames into discrete physical shots,extract features from video shots,and scene structure construction based on shots.In this paper,two crucial algorithms of video hierarchical structure mining ,shot clustering based on multi-features and scene change detection based on shot,are proposed.Our experimental results demonstrate the performance of scene change detection based on shot is better than that of shot clustering based on multi-features.

关 键 词：视频层次结构 视频结构挖掘 聚类分析 场景边界检测 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]