A new approach for high level video structuring
Yong-Moo Kwon, Chang-Jun Song, Ig-Jae Kim · 2002
The paper introduces a high level video structuring scheme using both color and motion features. Until now, color information has been used for video scene segmentation research. However, action movies with rapid motion of objects have characteristics of dynamic changes of color information. In this case, the traditional approach is not so appropriate. To overcome this drawback, a new criterion for video shot similarity using color and motion features is presented and discussed. Then, a new adaptive scene segmentation algorithm is proposed, of which the main features include adaptive weighting of color and motion similarity between two shots. Moreover, the improved overlapping links scheme is proposed for reducing the time cost for shot grouping. The proposed scheme is evaluated using the movie: Terminator II, together with implementing a prototype video browsing system using the structured video.