摘要
微表情时空特征描述子提取到的特征向量维数高,导致分类算法运算复杂度高,运行时间长,识别准确率较低。为此提出基于Fisher特征选择的方法,挑选鉴别力强的特征分量,对特征向量进行降维。采用“留一交叉验证”法,在CASMEII和SMIC两个数据集下分类实验表明,经Fisher特征选择后微表情的识别率可以达到75.71%和75%,分别较原特征向量识别率提高了61.22%和46.43%,而维数仅为原特征向量维数的4.18%和4.71%,分类所需时间是原方法的4.27%和1.61%。
The high dimension of feature vectors extracted from spatio-temporal descriptors of micro-expression results in high computational complexity,long running time and low recognition accuracy of classification algorithms.This paper proposes a method based on Fisher feature selection to select feature components with strong discrimination and reduce the dimension of feature vectors.By using the method of"leave-one-out-cross-validation",the classification experiments on CASMEII and SMIC datasets show that the recognition rate of micro-expression after Fisher feature selection can reach 75.71%and 75%,which are 61.22%and 46.43%higher than that of the original feature vectors respectively.The dimension is only 4.18%and 4.71%of the original feature vectors dimension,and the time required for classification is only 4.27%and 1.61%of the original method.
作者
张延良
卢冰
蒋涵笑
洪晓鹏
赵国英
张伟涛
Zhang Yanliang;Lu Bing;Jiang Hanxiao;Hong Xiaopeng;Zhao Guoying;Zhang Weitao(School of Physics and Electronic Information,Henan Polytechnic University,Jiaozuo 454150,Henan,China;School of Electronic and Information Engineering,Xi’an Jiaotong University,Xi’an 710049,Shaanxi,China;Center for Machine Vision and Signal Analysis,University of Oulu,Oulu FI-90014,Finland;School of Electronic Engineering,Xidian University,Xi’an 710071,Shaanxi,China)
出处
《计算机应用与软件》
北大核心
2020年第9期68-74,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61571339)
网络与交换技术国家重点实验室开放课题(SKLNST-2016-1-02)
河南理工大学博士基金项目(B2017-55)。
关键词
微表情识别
特征描述子
Fisher特征选择
识别准确率
Micro-expression recognition
Feature descriptors
Fisher feature selection
Recognition accuracy
