亮点文章

当前位置 >>亮点文章 >>亮点文章

Optimize x265 Rate Control: An Exploration of Lookahead in Frame Bit Allocation and Slice Type Decision时间: 2019-05-01 点击: 19 次

We present a rate control algorithm in which the frame base QP is determined with not only the coding complexity history but also the complexity changes and the data dependencies between the current and the near future pictures by exploring lookahead. Moreover, the quantization scale is introduced to the threshold specification in slice type decision, which identifies more pictures as B-type properly when increasing QP.

To improve the rate-distortion (R-D) quality,x265 rate-control makes a variety of vital decisions—such asscene cut detection, slice type decision, and coding-unit quantizationparameter (QP) offsets—leveraging on lookahead toevaluate the information propagation through the current and thenear future consecutive frames. However, as the frame base QPthat dominates the bit amount allocated to one frame was onlydetermined by the long-term complexity history in the originalalgorithm, the frame bit allocation became insensitive to therecent scene changes with the growth of coding time. In addition,the specified threshold in slice type decision, which was comparedto the estimated frame coding costs to detect the B-type slice,did not consider the impacts of quantization. As mentionedearlier, the irrational elements degraded the rate accuracy andthe R-D performance of x265. In this paper, the frame baseQP is determined with not only the coding complexity historybut also the complexity changes and the data dependenciesbetween the current and the near future pictures by exploringlookahead. Moreover, the quantization scale is introduced to thethreshold specification in slice type decision, which identifies morepictures as B-type properly when increasing QP. The proposedalgorithms were conducted in x265 version 2.4. Experimentsrevealed that under the default preset (–preset medium), 0.617 dBon average and up to 1.705 dB Bjøntegaard-Delta quality gainswere achieved, while saving the encoding time by 1.13% andimproving the rate accuracy by 4.2% on average.

上一篇:Regret Minimization for Reinforcement Learning by Evaluating the Optimal Bias Function 下一篇:Range Scaling Global U-Net for Perceptual Image Enhancement on Mobile Devices 返回列表

用户登录

用户注册