Human Identification Based on Deep Feature and Transfer Learning

Biometric based identity authentication has attracted much attention due to its unique advantages. Among all the biometric which can be used for authentication, human face based methods have been the most popular research area in both identity authentication and recognition. However, traditional method may result in poor performance when conducting face recognition under uncontrolled environmental. Deep network provides a more proper way to extract distinctive features for face recognition, however the performance of most deep network is usually limited by the number of training samples. Accordingly, this paper proposes a deep convolutional neural network combining with the idea of transfer learning and sparse representation to combat the disadvantage of traditional CNN on small sample task while simplifying the computational complexity. Abundant experimental results in different database show that compared with traditional method, our proposed method achieves higher and promising recognition rate.