In current video coding solutions, such as MPEG-1,2,4 Video or ITU-T H.26x standards, the encoder estimates the motion in inter frames (P and B frames) and transmits the motion vectors and the residue to the decoder. Thus, temporal correlations between frames are exploited and compression is achieved.
Due to block-based motion estimation, accurate compensation at object borders can only be achieved with small block sizes. However, the smaller the block, the more motion vectors have to be transmitted, resulting in a discrepancy to bit rate reduction. Therefore, the block size has a significant impact on compression performance and is limited to 4x4 pixel in H.264 / MPEG-4 AVC.
To overcome the drawbacks of segmentation and block-based motion compensation, motion estimation can be performed at the decoder. In this case, the predicted motion is already known at the decoder and thus, the transmission of the motion vectors can be omitted. Depending on the compensation accuracy, additional bits can be saved in the residue.
