独立成分分析在表情识别中的应用

The Application of Independent Component Analysis in Facial Expression Recognition

独立成分分析(ICA)是一种基于信号高阶统计特性的分析方法,本文尝试将这种方法应用于人脸表情的特征提取。首先对预处理后的图像用FastICA算法计算得到解混矩阵以及此训练样本集的影像独立基成分,然后利用影像独立基来构造一个投影空间,最后利用待识别的表情图像在这个空间上作空间影射,所得到得投影系数用以实现分类。为了减少运算量,本文研究了降维的训练样本集的独立成分分析。

ICA is an efficient method for the analysis of high order signals. In this article, the feature of facial expression is extracted by the application of ICA. Firstly, FastICA algorithm is used for computing decomposing matrix and independent components after necessary preprocessing is done. Then, we can use these independent components to construct a projection subspace. Finally, test sample is projected to the subspace and projection coefficients are obtained and computer can use these coefficients to recognize which expression it belongs to. In order to reduce the calculation task of computer, size reduction of training data is also studied in this paper.