Data partitioning based on motion estimation for robust video communications

In this paper, we present a novel data partitioning method, based on motion estimation, aiming at preventing error propagation when transmitting video streaming over wireless networks. Macroblock is treated as basic unit in this method, and all macroblocks; of a coded frame are differentiated into different levels of importance according to their own impact factor which is defined based on combination of two types of impact: one is the impact of a macroblock on the quality of next frame and the other is the impact on the quality of current frame. The first impact is evaluated by counting the total of referred times of all pixel within the macroblock and the second one is evaluated by calculating difference between original and predicted macroblock. Then, combined with header information, all macroblocks of a coded frame is partitioned into three types of partitions. Finally, different rates of FEC channel code are applied to different partitions such that the important data can be protected strongly. While improving the quality of video like other data partitioning methods, an advantage of the proposed method is to balance the quality of the current and the next frame. Compared with other data partitioning methods such as in H.263++ and H.264, the proposed method can efficiently limit error and prevent error propagation. Experimental results show the proposed method has obtained stable quality for entire video sequence and achieve better PSNR under the two-state Markov channel model.