UNIVERSITY OF CENTRAL FLORIDA AT TRECVID 2003
Yun Zhai, Xiaochun Chao, Zhang Yunjun, Omar Javed, Alper Yılmaz, Fahd Rafi, Saad Ali, Orkun Alatas, Saad M. Khan, Mubarak A. Shah · 2003
In this paper, we describe the methods for shotboundary detection, story segmentation, and feature extraction developed at the Computer Vision Laboratory, UCF, for TRECVID 2003 forum. To detect shot boundaries in videos, we use a multiresolution color histogram intersection technique in a coarse-to-fine fashion. Detected shot transitions are further classified into abrupt and gradual transitions. We have also developed a method to segment video clips into stories. Our method is based on only visual cues and is useful to separate commercials from the news stories. In addition, we have contributed to the feature extraction task and have developed methods to detect two features, namely, Non-Studio Settings and Weather News. The feature detection also relies on the visual information and exploits the structure of the shot transitions and the common annotations in the news videos. We present our results and its evaluation released by NIST. 1.