Successive refinement of motion compensated interpolation for transform-domain distributed video coding

In distributed video coding, the estimation of the side information at the decoder plays a key role in the final rate-distortion performance of the codec. The side information is commonly generated by motion-compensated temporal interpolation of the neighboring reference frames. In this paper, we propose a successive refinement after the decoding of each DCT subband to improve the accuracy of motion compensation between reference frames, in order to obtain a new side information estimation closer to the original Wyner-Ziv frame. The experimental results show that the proposed scheme can achieve up to 0.9 dB of improvement in rate-distortion performance for a GOP size of 2 and 2.4 dB for a GOP size of 8 for sequences containing high motion with respect to state-of-the-art techniques.