Everyday, we meet a vast amount of multimedia data in various areas such as communication, internet, broadcasting, educational and medical data, etc. But, with these large amount of data, it was not easy to find an appropriate datum by one’s request, To find data upon user’s query, each multimedia datum should be indexed and located. Above all, video indexing is an important technique because video is the most important information among various multimedia. But traditional indexing methods such as text-indexing is known to be inefficient for video data. Therefore, exact feature extraction and consequently efficient indexing techniques such as a content-based indexing technique is increasingly required recently for video data.
In this paper, we studied compressed domain-video indexing using object motion histogram for efficiency. For indexing, proposed feature is object motion(OM). Because OM is one of very important features of video sequence, one can obtain a cue about OM’s moving process in video data. At first, video data are segmented into shots, i.e, basic units for indexing and retrieval. Shot is defined as a continuous sequence of video frames which shows no significant inter-frame difference in terms of visual contents. Secondly OM is extracted by removing background motion from global motion in shot and then, clustered. Each OM has unique identification in a shot. Clustered OM has magnitude and direction. We adopted polar coordinate to describe OM because polar coordinate is suitable to describe the motion vector. To index OM, we divide each motion in polar coordinate into i number of magnitude and j number of direction so that $i\timesj$ motion types are obtained for various motions. Object motion histogram in a shot(OMHS) is obtained based on the polar division technique mentioned above. This histogram has three dimensions that consist of motion type, motion number and time or frame number in a shot. First moment of OMHS is used as a descriptor of OM.
To evaluate indexing performance of our proposed indexing technique, a video database having several MPEG-1 video sequence in MPEG-7 test set is constructed. Computer simulation and experiments show that the proposed indexing method is efficient and reliable for content-based video indexing.
