基于DenseNet+FPN网络的视频帧内CU快速划分算法  

Video frame-level CU rapid partitioning algorithm based on DenseNet+FPN network

作　　者：程栋梁 张吟龙 周道先 冯选 CHENG Dongliang;ZHANG Yinlong;ZHOU Daoxian;FENG Xuan(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang,Liaoning 110168,China;State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang,Liaoning 110016,China;Key Laboratory of Networked Control Systems,Chinese Academy of Sciences,Shenyang,Liaoning 110016,China;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang,Liaoning 110169,China;College of Information Engineering,Shenyang University of Chemical Technology,Shenyang,Liaoning 110142,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,辽宁沈阳110168 [2]中国科学院沈阳自动化研究所机器人学国家重点实验室,辽宁沈阳110016 [3]中国科学院网络化控制系统重点实验室,辽宁沈阳110016 [4]中国科学院机器人与智能制造创新研究院,辽宁沈阳110169 [5]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《光电子．激光》2024年第11期1208-1214,共7页Journal of Optoelectronics·Laser

摘　　要：针对多功能视频编码(versatile video coding,VVC)帧内编码中编码单元(coding units,CU)划分存在计算复杂度过高的问题,本文提出了一种基于DenseNet+FPN(feature pyramid network)的CU快速划分算法。该算法能够大幅度降低VVC的编码复杂度,减少编码时间。首先,本文提出了一种基于纹理复杂度的CU分类算法,来评估CU块的纹理复杂度。其次,提出一种基于DenseNet+FPN的网络模型,利用多尺度信息来优化CU划分,以适应多尺度情况下的编码需求。最后,设计了一个新的自适应的质量复杂度均衡损失函数,用于平衡编码质量和计算复杂度。所提算法进行了大量的实验分析,结果证明,与公共参考软件(WC test model 10.0,VTM10.0)相比,所提算法的帧内编码平均时间减少了44.268%,而BDBR(bj?ntegaard delta bit rate)仅增加了0.94%。To address the issue of high computational complexity in coding units(CU)partitioning for versatile video coding(VVC)intra-frame coding,this paper proposes a CU fast partitioning algorithm based on DenseNet+FPN(feature pyramid network).The algorithm significantly reduces the encoding complexity of VVC,resulting in reduced encoding time.Firstly,a CU classification algorithm based on texture complexity is proposed to evaluate the texture complexity of CU blocks.Secondly,a network model based on DenseNet+FPN is introduced,utilizing multi-scale information to optimize CU partitioning to adapt to encoding requirements in various scales.Lastly,a novel adaptive quality-complexity balanced loss function is designed to balance encoding quality and computational complexity.Extensive experimental analysis is conducted for the proposed algorithm,and the results demonstrate that compared to VVC test model(VTM)10.0,the average encoding time of the proposed algorithm is reduced by 44.268%,while the bjntegaard delta bit rate(BDBR)only increases by 0.94%.

关 键 词：H.266/VVC 帧内编码 纹理复杂度 快速划分算法 

分 类 号：TN919.8[电子电信—通信与信息系统]