An object-based segmentation approach for very low bit rate video sequence

An object based spatial-temporal segmentation approach for very low bit rate video sequences is proposed. Temporal segmentation is processed by the absolute displaced frame difference instead of the change detection mask. A fast rain-falling watershed algorithm is used for the spatial segmentation. The final result is obtained by the combination of the temporal segmentation and the spatial segmentation. The proposed method achieves excellent performance with MPEG-4 test sequences of videoconferencing.