Decoder side candidate motion vector selection for B-frames in H.264 video coding

In H.264 standard, because of differential coding of motion vectors, sudden motion changes in object boundaries increase the amount of bits used for coding especially at low bitrate regimes. In this paper, we propose a novel algorithm that uses decoder support for efficient coding of motion vectors in B-frames. In this method, decoder uses true motion estimation to determine a vector set that contains a limited number of candidate vectors. To improve the motion estimation accuracy, a constrained search is performed around each candidate vector. Vectors with low probability of being optimal are excluded from this search in order to reduce the size of candidate vector subsets. As a result since these candidate vector sets are fully determined by the decoder for each macroblock, there is no need for explicit coding of motion information, which reduces the bitrate required for coding. The algorithm provides an average 0.39 dB PSNR gain for B-frames when compared to the reference H.264 results at equal bitrates. Simulation results also indicate significant improvement in visual quality of the compressed B-frames.