Adaptive partition size temporal error concealment for H.264 using weighted double-sided EBME minimization

In existing temporal error concealment (TEC) methods for H.264, the partition size of the lost macroblock (MB) is often decided before estimating the motion vectors and there is no actual quality comparison between different partition modes. In this paper, we propose to select the best partition mode by minimizing the weighted double-sided external boundary matching error (WDS-EBME), which jointly measures the inter-MB boundary discontinuity, inter-partition boundary discontinuity and intra-partition block artifacts in the recovered MB. The proposed method estimates the best motion vectors for each of the candidate partition modes, calculates the overall WDS-EBME values for them, and selects the partition mode with the smallest overall WDS-EBME to recover the lost MB. We also propose a progressive concealment order for the 4×4 partition mode. Test results show that the adaptive partition size method always outperforms the fixed partition size methods with low complexity overhead. Both the adaptive and fixed partition size methods are much superior to the TEC method in the H.264 reference software.