计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (35): 190-194.

• 工程与应用 • 上一篇    下一篇

基于文本及视音频多模态信息的新闻分割

刘 扬1,郑逢斌1,樊卞玲2   

  1. 1.河南大学 计算机与信息工程学院,河南 开封 475001
    2.平煤集团开封东大化工有限公司,河南 开封 475003
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-12-11 发布日期:2007-12-11
  • 通讯作者: 刘 扬

TV news automatic segmentation base on text and audio-visual multi-modal features information

LIU Yang1,ZHENG Feng-bin1,FAN Bian-ling2   

  1. 1.College of Computer Science and Information Engineering,Henan University,Kaifeng,Henan 475001,China
    2.Kaifeng Dongda Chemical Industry Co.,L td.,Pingmei Group,Kaifeng,Henan 475003,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-12-11 Published:2007-12-11
  • Contact: LIU Yang

摘要: 提出了一种融合文本和视音频多模态特征的电视新闻自动分割方案。该方案充分考虑各种媒体特征的特点,先用矢量模型和GMM对文本进行预分割,用语谱图和HMM对语音预分割、用改进的直方图和SVM分类器对视频进行预分割。然后在时间同步的基础上,使用复合策略用ANN对预分割的数据进行融合,从而获得具有一定语义内容的视频段。实验结果表明此方法的有效性,并且分割后的视频片段具备较完整的语义信息特征,避免了分割的过度细碎的弊端。

关键词: 基于内容的检索, 多模态, 直方图, 语谱图, 视频分割

Abstract: TV news automatic segmentation scheme of fuse text and audio-visual multi-modal features information is presented.Regarding the different of all kinds of media feature,texts first were preparation segmented by GMM using vector model.Then audio are preparation segmented by HMM using spectrogram,and video were preparation segmented by SVM using improved histogram.At last,audio-visual and text segmentation are integrated to ANN base on synchronization and with compound strategy to get segmentation of the video with respect to its semantic meaning.The experimental results show the approach is valid,and avoids the problem of a far too segmentation of the video.

Key words: content-based retrieval, multi-modal classifiers, histogram, spectrogram, video segmentation