基于双层虚拟视图和支持向量的人脸识别方法

Face Recognition Based on Two-Layer Generated Virtual Data for SVM

针对训练样本较少情况下的人脸识别问题,该文提出基于生成视图和支持向量机的识别方法.在人脸识别的实际应用中,处理的人脸图像,每类往往只有很少的样本,以至于不能充分表达样本的实际分布,需要对训练样本的数据进行有效地扩充.为此首先通过对人脸图像中眼睛中心位置的扰动,利用面像模板,自动生成该人脸的多个虚拟人脸图像,并与原图像一起形成第一层的人脸库,然后应用 Eigenface方法得到人脸的特征数据,按照每个类的样本数据分布,应用内插法和外推法进行第二层次的扩充.在 ICT YCNC和 UMIST人脸库中应用 Multi Class支持向量机对得到的数据进行实验,结果表明,在样本不足的条件下利用支持向量识别人脸,生成虚拟视图是一种有效的方法.

This paper presents support vector machines (SVM) for few samples-based face recognition with two-layer generated virtual training data. The few samples cannot express all the conditions of the test data, such as the changes of poses, illuminations etc.. Authors generalize the samples and the data to other conditions: first, corresponding to the original face images, moving the locations of the eyes center on the face images and using the face mask template, different normal face images can be gotten; second is to the feature vectors of the face images, authors use principal component analysis to extract the features of the face images, and then to any class of the face images, they use linear interpolation and extrapolation methods to extend new data according to the distribute of the feature vectors. After the data drawn out use SVM to train and test. In the ICT-YCNC and UMIST face data base, the proposed system obtains competitive results, and shows efficiency of the method.