Fast Rate Estimation for H.264/AVC CAVLC Encoder Based on Parallel Processing

The entropy engine of context adaptive variable-length coding (CAVLC) has been employed in the H.264/AVC baseline for its high coding efficiency. However, the related complexity is also high owing to the utilization of inter-symbol dependency in CAVLC. In this paper, an efficient and accurate bit-rate estimation scheme is proposed to avoid the actual CAVLC encoding in RDO for mode decision and to meet the requirement of low-complexity video coding implementations. The proposed approach firstly parses the characteristics of syntax elements. Then guided by the principle of CAVLC, a fast and precise algorithm is designed for accurate estimation of the bit-rate in rate-distortion mode decision. The structure of the proposed algorithm is well suited to parallel processing for hardware implementation such as chip design regarding H.264/AVC, which is beneficial to further reduce the encoding time for video coding applications. When the proposed method is incorporated into reference software of H.264/AVC in a sequential order, a reduction about 62% in the CAVLC encoding time in RDO can be observed with negligible degradation in the rate-distortion performance.