Residual Syntax Elements Analysis and Design Targeting High-Throughput HEVC CABAC

摘要

Video processing is a topic of major interest due to the demand for real-time delivery of increasing resolution videos. Moreover, Internet traffic related to high-quality video processing tends to be the dominant type of information transmitted. The high-efficiency video coding (HEVC) standard is a suitable and efficient choice to overcome the challenges regarding the matter presented. In HEVC, the final encoding step is the entropy encoding, in which the context-adaptive binary arithmetic coding (CABAC) algorithm is the sole option used. Many recent research efforts have been successful in enhancing CABAC throughput for a hardware approach, which is a remarkable difficulty due to the intrinsic dependencies generated by the algorithm. The throughput improvement happens due to the capability of these entropy-encoding designs to consume multiple input data coming from the previous encoding steps. Nevertheless, the same care in terms of data generation shall occur for the stage right before the CABAC block. Otherwise, a transfer of the potential processing bottleneck would move to the input data fabrication. Therefore, this paper proposes a novel scheme, named multi residual syntax element treatment (MRSET), to be compliant with these recent high-throughput CABAC designs, being able to supply the entropy coding with enough input data and thus it would not starve. A complete analysis involving HEVC reference software and recommend video sequences occurred to drive MRSET inception. Different multi-core MRSET hardware architectures results are provided, and the best suitable option is defined based on the characteristics of the state-of-the-art CABAC architectures.