Temporal error concealment based on content adaptive block matching

It is essential to accurately estimate the motion vectors of the corrupted macroblocks in temporal error concealment. Usually the lost motion vector is estimated by searching the motion vector of the surrounded domain block in the previous frame. Ideally image content should be taken into account when selecting the pixels used for matching to improve the accuracy of the estimation. However, traditional methods are not aware of the image content. A content adaptive block matching algorithm is proposed in this paper. In our method, the size of the domain block is adaptive to the image content. Furthermore, since the edges in an image can reflect its content to some extent, the edge pixels in the domain block are selected for motion estimation when the domain block is textually complex. Nevertheless more pixels besides the edges are selected for a smooth domain block. The accuracy and performance of the proposed EC method is validated experimentally. Comparative evaluation with existing EC methods demonstrates the proposed method is superior to the others in both objective and subjective retrieval video quality.