Key frame selection by macroblock type and motion vector analysis

Key frame selection is the process to select the most representative frames of a long video sequence. The paper describes a new algorithm for determining key frames in a video shot. We use macroblock (MB) type and motion vector information of MPEG compressed video to determine the local maxima of visual content change in a shot, i.e. the frames that are most noticeable to the viewer. This algorithm exploits the important cue of the intensity of motion activity and visual content change within a shot from the corresponding MB type and motion vector characteristics. The advantages of this algorithm are the efficient extraction of the required information in the MPEG compressed domain with low complexity VLC decoding and high accuracy. Results show that the proposed algorithm can successfully select the key frames that are sufficient to represent a shot.