Content-based video retrieval using motion descriptors extracted from compressed domain

Video content description has become an important task with the standardization effort of MPEG-7, which aims at easy and efficient access to visual information. In this paper we propose a system to extract the features from compressed MPEG video based on the motion vector information. The global features like motion activity and camera motion parameters are extracted from the decoded motion vectors and the object features such as speed, area and trajectory are obtained after the object segmentation stage. The number of objects in a given video shot is determined by the proposed K-means clustering algorithm and the object segmentation is done by applying EM algorithm.