The Myth of Semantic NEVENKA Phdlps Video Retrieval DIMITROVA 345 Scarborough Road, Brtarcllff Manor, New York Laboratories, The mission of this paper is to draw attention to some problems in current research directions in automatic digital video retrieval and suggest integrated retrieval based on image, motion, and audio analysis as a promising direction. Initially researchers concentrated on the image component of digital video analysis. However, as I will explain in the following, there are inherent problems in treating video as a sequence of images, since motion and activities cannot be inferred from selected and isolated frames only. In addition, the audio track of the video clip is almost ignored in the analysis. Current trends in video analysis and content retrieval based on semantics suggest a great deal of enthusiasm on the part of researchers. Recent projects include video segmentation, extraction of representative key frames, video annotation, parsing of TV news, and parsing soccer games [ICMCS 1995]. These projects rely primarily on the current state of the art in computer vision and imageprocessing technology. However, the fact that we have fast machines today, as well as an abundance of algorithms available as a result of forty years research
/lp/association-for-computing-machinery/the-myth-of-semantic-video-retrieval-2KF3gDAGxR