微表情识别中面部动力谱特征提取的PCA改进

PCA-based Modification of Dynamics Map Feature Extraction in Micro-Expression Recognition

针对用于人脸微表情识别的面部动力谱特征(FDM)提取方法中迭代算法抽取时空立方体主方向时时间复杂度高的问题,提出基于主元分析(PCA)的改进算法。首先将数据中心平移到原点,其次估计相关矩阵,最后估计其特征向量和特征值,取最大特征值对应的特征向量方向作为时空立方体的主方向。采用Oulu大学SMIC微表情数据库中的微表情片段作为实验样本,选择支持向量机(SVM)作为分类器,对改进算法和原算法进行人脸微表情识别对比实验。结果表明,两种算法识别率相近,但改进算法在计算时间上远短于原算法。可见,改进算法在准确找出时空立方体主方向的同时,能大大降低原算法的计算复杂度。

Aiming at high time-complexity problem for iterative algorithm to extract the main direction of spatiotemporal cuboids in FDM(Facial Dynamics Map)method for facial micro-expression recognition,a modified FDM algorithm based on PCA(Principal Component Analysis)is proposed.Firstly the data center is moved to its origin,then the correlation matrix estimated,and then the eigenvector and eigenvalue calculated.The direction of eigenvector corresponding to maximum eigenvalue is taken as the main direction of spatiotemporal cuboids.The micro-expression fragments in the SMIC micro-expression database of Oulu University are employed as experimental samples,and SVM(Support Vector Machine)is selected as the classifier,the comparison of modified algorithm with original algorithm is done for facial micro-expression recognition.The experiment results indicate that the recognition rates of the two algorithms are similar,but the modified algorithm is much shorter than the original algorithm in computing time.It can be seen that the modified algorithm could greatly reduce the computational complexity of original algorithm while accurately estimating the main direction of spatiotemporal cuboids.