電視新聞及廣告分段與視訊短片搜尋之研究

Abstract

本論文旨在發展自動區分電視新聞中的廣告段落, 以及能準確搜尋目標短片之技術. 在論文中, 我們以聲音及視訊之特徵差異區分新聞及廣告, 並以新聞節目結構上之特性, 提升分辨新聞及廣告的準確性; 另外, 我們發展出以鏡頭為準 (shot-based) 之短片搜尋技術, 以利加強於實際的廣播節目錄影中搜尋特定視訊短片之準確性.

This paper mainly develops the technique of dividing commercials from TV news programs, and the technique of video clip searching. We use the audio and video features to distinguish between news and commercials, and then some properties of the structure of TV news programs are applied to improve the accuracy of news and commercial segmentation. And, we also develop a shot-based video clip search method to enhance the accuracy of video clip searching in the practical broadcasts.