Optimal Deep Rate Control for Intra Coding in High-Efficiency Video Coding Standard

This paper proposes a novel Optimal Deep Rate Controller (ODRC) designed for intra-coding configuration of the High-Efficiency Video Coding standard. The ODRC incorporates a Convolutional Neural Network-based Rate-Quantization Model (CRQM) to effectively predict bit consumption across the entire Quantization Parameter (QP) range at the Coding Tree Unit (CTU) level. The proposed rate controller employs an optimization algorithm to minimize the buffering delay required for video communications. By establishing a specific search space through the CRQM, a greedy search algorithm is utilized to determine the optimal frame-level QP, thereby minimizing discrepancies between buffer occupancy and target occupancy. Unlike CTU-level rate controllers, which can introduce quality variations due to QP fluctuations among CTUs, the frame-level ODRC maintains consistent objective quality across CTUs within a frame. The ODRC is integrated within the standard reference software HM-۱۶.۲۰. Comparative evaluations with the default rate controller, RC-HM, in the same software, demonstrate the superior performance of ODRC in terms of both delay and bit error ratio. Experimental results indicate that ODRC achieves a notably lower average buffering delay of ۰.۰۲s and a lower bit error ratio of ۱۱.۲۵%, in contrast to RC-HM's ۰.۳s and ۴۴.۷۲%, respectively, emphasizing its effectiveness for HEVC low-delay applications.

کلیدواژه ها:

Convolutional Neural Network (CNN) ، High-Efficiency Video Coding (HEVC) ، R-Q Model ، rate control