摘要
人脸表情识别是指利用计算机技术、图像处理、机器视觉等技术对人脸表情图像或图像序列进行特征提取、建模,以及表情分类的过程,从而使得计算机程序能够依据人的脸部表情信息推断人的心理状态。人脸表情识别主要分为三个阶段:人脸检测、表情特征提取、表情特征分类。其中,表情特征的选取是人脸表情识别的关键步骤,特征选取的好坏直接影响表情分类的效果。论文提出了一种基于IMF解析信号能量熵的人脸表情特征提取方法,将希尔伯特黄变换方法应用到人脸表情识别中。首先,对表情图像进行Radon变换,得到人脸表情信号,然后对该信号进行经验模态分解(EMD),得到一系列本征模态函数(IMF),对得到本征模态函数(IMF)进行Hilbert变换,得到IMF解析信号,计算瞬时振幅,瞬时频率。选择IMF以及其解析信号的振幅作为特征向量,计算其能量判别熵,选择同类之间有较小判别熵,不同信号类之间有较大判别熵的特征作为表情分类的特征向量。采用PCA算法对选取的特征进行降维,使用支持向量机(SVM)对两类表情进行分类。
Facial emotion or facial expression recognition refers to using computer technology,image processing and machine vision technology to process the object from a given image or video sequence for feature extraction,modeling,classification to identify the psychological mood of the subject.Facial expression recognition is mainly divided into three stages,including face detection,face feature extraction and expression classification.Expression feature extraction and selection is a key step in efficient and effective facial emotion recognition and may affect the classification results.In this study,a novel approach of face expression feature extraction is proposed based on energy entropy of IMF analytic signal.Firstly,a Radon transform is made for the facial image to obtain a facial signal and the facial signal is decomposed into a number of IMFs,using the EMD algorithm.Then,with the Hilbert transform for etch IMF,the amplitude of IMF analytic signal can be acquired.In this work,the element which has larger discriminant entropy between the different classes and smaller discriminant entropy is chosen in the same class as the feature vector for emotion classification.Principal Component Analysis(PCA)is independently applied on features extraction for dimensionality reduction.These dimensionality reduced features are fed to the Support Vector machine(SVM)classifiers for classification.
出处
《计算机与数字工程》
2016年第3期529-532,共4页
Computer & Digital Engineering
基金
支持群体交互的大规模虚拟环境构建技术及系统(编号:2013AA013902)资助