结合鲁棒PCA特征与随机森林的表情识别方法  

Facial expression recognition method combined with robust PCA feature and random forest

作　　者：欧中亚[1,2] 山田宏尚[2] OU Zhong-ya;YAMADA Hironao(Department of Information Management,Henan Institute of Economic and Trade,Zhengzhou 450046,China;Mechanical System Faculty of Engineering,Gifu University,Nagoya 5011193,Japan)

机构地区：[1]河南经贸职业学院信息管理系,河南郑州450046 [2]日本岐阜大学工学研究科机械系统,岐阜名古屋5011193

出　　处：《计算机工程与设计》2018年第2期580-584,595,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61063028)

摘　　要：为提高表情识别的识别率,提出一种鲁棒的PCA特征提取方法,结合随机森林学习方法实现人脸表情的识别。该方法主要包括图像预处理、表情特征提取和表情特征的训练与分类3个部分,其主要创新在于鲁棒的PCA特征提取方法。融合欧氏距离和明氏距离两种距离计算方法求取样本均值,采用梯度下降算法迭代寻找最优的样本中心和投影矩阵,提取适应不同样本的鲁棒PCA特征;在图像预处理阶段提出改进的Gamma校正方法,避免在光照校正时大幅改变图像的整体亮度分布。实验结果表明,该方法对表情的识别率高,运算效率高。To improve the recognition of facial expression recognition,a robust PCA(principal components analysis)feature extraction method was proposed,to realize facial expression recognition by combining random forest learning method.This method included three parts such as image processing,feature extraction and expression feature training and classification.The main innovation lied in robust PCA feature extraction method.Euclidean distance and Minkowski distance were fused to calculate sample mean,and gradient descent algorithm was used to find the optimal sample center and projection matrix iteratively,for extracting robust PCA feature for different samples.A modified Gamma correction method was introduced in the process of image preprocessing,to avoid substantially changing the overall lightness of the image distribution on the process of illumination correction.Experimental results show that the proposed method has high recognition rate and efficiency for facial expression recognition.

关 键 词：主成分分析 表情识别 随机森林 GAMMA校正 欧氏距离 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]