综合利用声视特征的新闻视频结构化模型  

A News Video Model Based on Audio and Video Features

作　　者：史迎春[1] 方鹏飞[2] 周献中[1] 

机构地区：[1]南京理工大学自动化系,南京210094 [2]东南大学计算机科学与工程系,南京210094

出　　处：《计算机工程与应用》2004年第32期99-101,189,共4页Computer Engineering and Applications

摘　　要：视频结构化表征和基于这种表征进行相似度比较是视频检索的前提和最基础的重要工作。该文受视频制作过程的启发,通过分析新闻视频的内容结构特点,提出一个基于多语义抽象层次表示的5层视频目录结构化模型;在此基础上,采用分层处理思想,通过综合利用声视特征实现了基于目录结构的新闻视频结构化,克服了单纯用视觉特征难以胜任场景分段的困难,实验验证了该文思想及相应算法的有效性。Structured expression and similarity comparison are the most important jobs in video retrieval.In the paper,a structured five-layer video catalog model is proposed based on multi-level abstract semantic,with the analysis of the process of video making and the structure features of news video.Then structurization of news video based on the cata-log is realized by using the idea of multi-layer processing and integrating audio and video features.Experiments show that the idea is reasonable,the algorithm is available and can achieve good results since it overcomes the difficulty in video semantic parsing when using only visual features.

关 键 词：视频检索 新闻视频结构化 视频分段 音频分段 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]