Abstract
We have already proposed the automatic summary extraction method from the MPEG coded audio-visual data, by using audio-visual characteristics on compressed domain. Our previous method assumes that the visual features such as motion are known a priori. In this paper, we show the extended method for extracting the audio-visual summary without any knowledge of the content by adaptively evaluating the visual features. Additionally, in order to extract summary, our method utilizes some descriptors defined in the MPEG-7 standard and can generate some MPEG-7 descriptions; our method is very flexible and can be easily applicable to the MPEG-7 applications.
