Various pose face recognition with one front training sample
单样本多姿态人脸识别研究

Almost all algorithms for face recognition have tight relationship with the images number of each person. The recognition rate increases with the increasing training number of each class. But in applications, it is not practical to ask for many training images from each person. A new method, which can generate the simulated images of face after rotating an angle, was proposed. It generalized the method of Fisherfaces and uncorrelated image projection discriminant analysis to one sample per person. The recognition rates of Principal Component Analysis (PCA), Fisherfaces, and Two dimension's PCA (2DPCA) were also studied. The experimental results on FERET face-databases indicate that after adding virtual images, the recognition rates increase greatly, and the best recognition rate has improved by 28.2%.