一种基于组合神经网络的HEVC帧间预测方法  

A HEVC interframe prediction method based on the combinatorial neural network

作　　者：韩强 吴帆[2] 蒋剑飞[1] HAN Qiang;WU Fan;JIANG Jian-fei(School of Microelectronics,Shanghai Jiaotong University,Shanghai 200240,China;China Shipbuilding Industry Corporation 711 Research Institute,Shanghai 201100,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]中国船舶重工集团公司第七一一研究所,上海201100

出　　处：《信息技术》2021年第4期1-5,10,共6页Information Technology

基　　金：国家重点研发计划项目(2018YFB2100100)。

摘　　要：高效视频编码(HEVC)作为最新视频编码标准,有着非常高的压缩效率,但是由于各种新技术的提出,其编码复杂度也大大提高。复杂度对视频编码有着重要意义,低复杂度编码的研究非常必要。利用神经网络进行HEVC的分区预测为低复杂度编码提供了有效的解决方案。文中提出了一种基于卷积神经网络(CNN)和长短期记忆网络(LSTM)的组合网络架构来对帧间分区进行预测的方法,利用自建数据库对网络进行训练;文中设计了一种预搜索模块来建立训练数据库,仿真结果表明,神经网络的精度可达87%,利用该网络架构进行帧间预测可以实现52%~71%的复杂度节省。As the latest video coding standard,High Efficiency Video Coding(HEVC)has a very high compression efficiency,however,with the introduction of various new technologies,its coding complexity has also been greatly increased accordingly.Complexity is of great significance to video coding,and research on low-complexity coding is very necessary.The partition prediction of HEVC using neural network provides an effective solution for low-complexity coding.This paper proposes a method to predict CU partitions at inter-modes,which is based on a combined network architecture of a Convolutional Neural Network(CNN)and a Long Short-Term Memory Network(LSTM)and we designed a Pre-search module showing that the accuracy could reach 87%,and it could save 52%to 71%complexity when doing interframe prediction.

关 键 词：低复杂度 帧间预测 长短期记忆网络 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]