摘要
提出基于DirectShow组件结构实时提取各类音视频特征的方法,并综合利用这些特征,通过对常用的视频关键帧提取算法进行分析比较,提出了基于FCNN模型的自适应关键帧提取方法,该方法可以适应于各种视频类型。以关键帧为基础,提出利用插值算法建立视频摘要的方法。最后以新闻视频为例,给出了基于内容的新闻视频摘要系统(KBVA-1)。
A real-time method of audio-visual feature extraction was introduced. Through the comparison of the present key-frame- extraction methods, a new algorithm, based on the FCNN model, was proposed. To effectively establish a video abstract, a method based on the key frames and their interpolation was introduced. As an example, a news video abstract prototype （KBVA-1） was developed using DirectShow architecture and COM criterion.
出处
《计算机工程与设计》
CSCD
北大核心
2005年第10期2598-2601,共4页
Computer Engineering and Design
基金
湖北省科技攻关基金项目(2004AA101C94)
关键词
基于内容
关键帧
视频摘要
content-based
keyframe
video abstract
