基于EM-PCA和级联分类器的人脸检测

Face detection based on EM-PCA and hierarchical classification

为了提高人脸检测的速度及鲁棒性,提出了一种基于级联分类器和期望最大、主成分分析(EM-PCA)的人脸检测方法.该方法在训练阶段利用不同分辨率的训练样本来训练2个fisher线性分类器,再利用EM-PCA提取特征来训练非线性支持向量机(SVM);在检测阶段,首先通过2个fisher线性分类器快速过滤掉大量的背景区域,再利用非线性支持向量机对余下的候选区域进行进一步验证,以确认是否为人脸.实验结果证明了该方法的有效性和正确性.

In order to improve the speed and robust of detecting human face,an algorithm of face detection based on EM-PCA and hierarchical classification is presented.In the training step,different resolutions of train samples are used for training two kinds of fisher linear discrimination,and train the nonlinear SVM by using the feature extracted by means of EM-PCA.In the detection step,the fisher linear discrimination is used for excluding large parts of backgrounds,and use the SVM to perform the final detection.Experimental results show that the new algorithm is feasible and efficient.