Abstract:

As one of the contactless biometric methods, face recognition is widely used in many situations. However, the traditional face recognition method can not meet the current demand because of its low recognition accuracy and limited application in many occasions. In this paper, we proposed a deep learning method to achieve facial marker detection and unrestricted face recognition. In order to solved the problem of face marker detection, this paper proposed a layer-by-layer training method for deep convolutional neural networks to help convolutional neural network convergence, and proposes a sample transformation method to avoided overfitting;In order to solve the face recognition problem, this paper proposed a SIAMESE convolutional neural network, which was trained on different parts and scales. Experimental tests showed that the accuracy of ORL and face recognition algorithms reached 91% and 81%, respectively.

Key words: facial Landmark detection, unrestricted face recognition, deep learning, convolutional neural network, SIAMESE network