Measuring the motion similarity in video indexing

Recent growth in the number of digital videos available motivates the development of video databases for the efficient management of these videos. Common video databases use image information calculated from key frames such as color, texture, shape to index videos, and only few of them are capable to store and retrieve motion information. In this paper we present a method for automatic motion based video indexing and retrieval. A prototype system has been developed, which automatically splits a video into shorter unit called shots, extract representative frames from each shot and estimates motion information in the neighborhood of the r-frame with the phase correlation method. A query can be an extracted motion information of a short sample image sequence, or a quantitatively given motion intensity in five region of the frame.