Motion vector recovery for error concealment based on distortion modeling

If channel errors are introduced during real-time video transmission and cannot be corrected properly at the receiver, we can apply error concealment techniques to repair damaged portions of the frame by exploiting spatial and temporal redundancies in the video signal. In general, compressed bitstreams generated by an MPEG video coder are quite sensitive to channel errors. If one packet is lost or received with errors, not only the current frame will be corrupted, but also the errors will propagate to succeeding frames. We define a new model to describe the distortion of corrupted macroblocks in the bitstreams generated by an MPEG-2 video codec. Based on the proposed distortion model, we have developed several error concealment algorithms to reduce the effect of channel errors.