True-Error Detection of Compressed Video: Temporal Exploration

In this paper, we propose a novel error detection method in the temporal domain for H.264/AVC encoded video streams. The corrupted macro blocks (MBs) are detected by exploiting the correlations between MBs in the neighboring two frames. Correlation coefficient and mean of residual block are introduced to quantify the correlations in the temporal domain. A supervised classifier based on probability density functions of proposed features is designed for error detection and expectation maximum algorithm is employed to find the optimal parameters of the classifier. Eight corrupted H.264/AVC encoded video sequences are used in our experimental work for test, and experimental results show that the proposed error detection method can detect the corrupted MBs effectively.