A graphical model for multimedia data indexing and retrieval

Spatio-temporal semantics is an important part of the content in video. The relative positions and movements of the objects of interest in a video segment constitute a crucial component of the information to be retrieved in a multimedia database system. This paper addresses this challenge and provides an efficient technique for video data retrieval using descriptions of relative object movements. Such descriptions are built based on a graphical representation, named Coordinate Valued Neighborhood Graph. This graphical model extends the functionality of Neighborhood Graph model from still images to video sequences. The resulting scheme is integrated to a content-based retrieval processing system, PICTURESQUE.