Single sample per person face recognition based on deep convolutional neural network
Junying Zeng, Xiaoxiao Zhao, Chuanbo Qin, Zuoyong Lin · 2017
As for that there is only one training sample in single sample per person(SSPP) face recognition(FR), deep learning(DL) method is very difficult to be used. To overcome this problem, the paper proposes a series of methods to make it possible to be used in SSPP FR. Firstly, an expanding sample method is proposed to increase training sample. Secondly, a learned Deep Convolutional Neural Network(DCNN) model which is trained with amounts of face images and can represent face very well is brought in. Then, these expanding samples are used to fine-tune the DCNN model. Thirdly, the fine-tuned DCNN model is used to perform experiment. Experiments demonstrate that a better performance is obtained by using the fine-tuned DCNN model.