面部表情交互识别方法研究  

Study on Facial Expression Interactive Recognition Method

作　　者：刘秋实 赵长宽 张昱 陈默 于戈 LIU Qiu-shi;ZHAO Chang-kuan;ZHANG Yu;CHEN Mo;YU Ge(School of Computer Science and Engineering,Northeastern University,Shenyang 110169,China)

机构地区：[1]东北大学计算机科学与工程学院,沈阳110169

出　　处：《小型微型计算机系统》2021年第10期2095-2100,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金联合基金项目(U1811261)资助。

摘　　要：人类表情是反映交互双方情感意图的重要社交信号,识别表情交互中隐含的人际关系行为模式是情感计算的一个重要问题.本文以尊敬服从、信任接受等8种人际关系行为模式为分类依据,选取典型交互场景下交互双方的的表情交互序列为研究对象,建立了基于长短期记忆神经网络的交互情感计算模型,完成表情交互序列分类.进而,引入共情这一概念,提出对其进行量化的方法,将其量化值作为新参数改进模型.对比实验表明,该方法具有较高的准确率,且开拓了人与人之间交互识别的新途径.Human expressions are important social signals that reflect the emotional intentions of the interacting parties.Identifying the behavior pattern of the interpersonal relationship from the interaction sequence of expressions is an important problem.The eight interpersonal relationship behavior patterns such as respect,obedience and trust acceptance are chosen as the classification standards to label data.The expression interaction sequences composed of key frame expressions in typical interaction scenes are constructed as the research object.And then an interactive emotion calculation model based on the long and short-term memory neural networks to complete expression interaction is proposed.Furthermore,the concept of empathy is introduced,and a method to quantify it is proposed,and its quantified value is used as a newparameter to improve the recognition model.The comparative experiments showthat our proposed method has a higher accuracy rate,and it explores a newway of interactive recognition.

关 键 词：人-人交互 表情交互序列 表情识别 神经网络 人际关系行为模式 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]