基于神经网络的HEVC帧间预测  

Neural network based inter prediction for HEVC

作　　者：王洋[1] 高文[1] WANG Yang;GAO Wen(School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2020年第3期386-391,共6页Intelligent Computer and Applications

基　　金：国家973项目(2015CB351804)。

摘　　要：高效视频编码标准HEVC是最新的数字视频编码标准,其中,帧间预测在去除时域相关性方面扮演重要角色。在传统的帧间预测方法中,因为仅仅利用了时域信息,所以帧间预测的准确性是有限的。在本文中,提出了利用时域-空域信息的基于神经网络的HEVC帧间预测算法。具体来讲,在本文提出的算法中,首先设计了一个包含全连接网络(fully connected netw ork,FCN)和卷积神经网络(convolutional neural network,CNN)的神经网络结构。其次,时域相邻像素和空域相邻像素输入到FCN网络,FCN网络输出的结果和当前预测块输入到CNN网络中,CNN网络生成最终的当前块的更准确的预测块。实验结果表明,与HEVC参考软件HM 16.9在通用测试条件Low Delay P(LDP)下编码,本文提出的算法能够带来平均1.7%(高达8.6%)的BD-rate节省。HEVC is the latest video coding standard,in which inter prediction plays an important role to reduce the temporal redundancy.The accuracy of inter prediction is limited since only temporal information is used in conventional algorithms.This paper proposes a neural network based inter prediction algorithm for HEVC by using the spatial-temporal information.In the proposed algorithm,first a neural network architecture is designed consisting of a fully connected network(FCN)and a convolutional neural network(CNN).Then the spatial neighboring pixels and the temporal neighboring pixels are inputted into FCN.The output of FCN and the prediction of current block are inputted into CNN,which will generate the more accurate prediction of current block.Experimental results demonstrate that the proposed method can achieve average 1.7%(up to 8.6%)BD-rate reduction in low delay P test condition compared to HM 16.9.

关 键 词：HEVC 帧间预测 神经网络 卷积神经网络 全连接网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]