基于CNN的λ域帧内码控最佳码率分配算法  

CNN based optimal rate allocation algorithm forλ-domain intra coding rate control

作　　者：林湧 杨郑龙 罗亦茜 刘欣昱 Lin Yong;Yang Zhenglong;Luo Yixi;Liu Xinyu(Dept.of Urban Rail Transit,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学城市轨道交通学院,上海201620

出　　处：《计算机时代》2023年第9期87-91,95,共6页Computer Era

基　　金：张江基金项目(No.ZJ2020-ZD-009)。

摘　　要：在λ域帧内码控中,提出一种基于卷积神经网络(Convolutional Neural Network,CNN)的帧内码控最佳码率分配算法。首先利用双曲线函数拟合编码树单元(Coding Tree Unit,CTU)的率失真(Rate Distortion,RD)特性。设计双分支卷积神经网络(Dual-Branch Convolutional Neural Network,DBCNN)预测率失真关键参数。然后根据帧级率失真优化(Rate Distortion Optimization,RDO),建立帧级目标码率与CTU码率分配等式关系,推导帧级拉格朗日参数λ。最后反演出最佳CTU码率分配。实验表明,该算法能够显著提高帧内码控编码性能,并具有较高码控精度。Inλ-domain intra coding rate control,an optimal intra coding rate allocation algorithm based on convolutional neural network(CNN)is proposed.Firstly,the rate-distortion(RD)characteristics of the coding tree unit(CTU)are fitted using the hyperbolic function.Dual-branch convolutional neural network(DBCNN)is designed to predict the key parameters of RD.Then,according to the frame-level rate-distortion optimization(RDO),the equation relationship between the frame-level target rate and the rate allocation of the CTU is established,and the frame-level Lagrange parameterλis derived.Finally,through program optimization,the optimal rate allocation of the CTU is performed to achieve the best balance between coding quality and code rate control.Experimental results show that the algorithm can significantly improve the coding performance of intra coding rate control,and has high rate control accuracy.

关 键 词：H.265/HEVC 帧内编码 卷积神经网络 λ域码率控制 率失真特性 

分 类 号：TN919.81[电子电信—通信与信息系统]