Spatio-temporal feature extraction from compressed video data

Meaningful feature extraction from video data is an important step in content-based video retrieval system. In particular, it is desirable to detect these features from compressed video data because this requires less processing overhead. Some approaches are proposed to extract features from compressed data by computing the variances of DCT coefficients and motion vectors. However these approaches do not consider objects in video data, so undesirable results are often generated. We propose a new method in extracting spatio-temporal features such as dominant regions, color information and motions from compressed video data for content-based video processing.