Efficient Variable Block Size Selection for AVC Low Bitrate Applications



The Advanced Video Coding (AVC) standard proposes the usage of Variable Block Size (VBS) motion-compensated prediction and mode decision aiming for an optimized Rate-Distortion (R-D) performance. Unlike Fixed Block Size (FBS) motion-compensated prediction, where all regions of the pictures are treated similarly in terms of temporal prediction, VBS increases the efficiency of encoding by allowing more active regions to be represented with more bits than less active ones. The main concern regarding the usage of VBS motion-compensated prediction is the dramatic increase it adds to the encoder computational requirements, which not only prevents the encoder from satisfying real-time constraints, but also makes it impractical for hardware implementation. This paper presents an efficient VBS selection scheme, which can be applied to any VBS Motion Estimation (ME) module, leading to significant reduction in its computational requirements with minor loss in the quality of the reconstructed picture. The computational requirements reduction is achieved by minimizing the number of required ME searches and simplifying the Mode Decision (MD) operation. In order to meet different applications’ demands, the proposed algorithm can be adjusted to function at any of three operating points, trading off computational requirements with R-D performance. In the paper, the algorithm is described in detail, focusing on the theoretical computational requirements savings. This theoretical analysis is then supported with simulation results performed on three benchmark video sequences with various types of motion. Keywords-H.264/AVC, motion estimation, variable block size.


Download from here

Reference: Ihab Amer, Wael Badawy, Graham Jullien, Adrian Chirila-Rus, Robert Turney, and Rana Hamed, “Efficient Variable Block Size Selection for AVC Low Bitrate Applications,” IARIA on-line journals, 2010 Vol. 1&2, July 2010.

Link to the list of other Peer Journal Publications

Leave a Reply