Face Recognition Based on Two-stage CNN Combined with Transfer Learning

In recent years, pattern recognition has received wide attention, especially face recognition. Research shows that traditional methods have limits in face recognition under unrestricted environment because of illumination, facial expression, pose and so on. Deep learning is chosen to solve these problems. In this paper, we propose a complete method for face recognition under unrestricted environment by deep learning. First, we design a novel two-stage convolutional neural network for face detection and alignment. The first stage of the network aims at searching face windows and their bounding box regression vector. The second stage merges facial candidates which are covered highly. Then, the network finishes the task of face detection and alignment whose outputs are face regions and the position of the five facial landmarks. It achieves great performance on angle of faces in image and obtains accuracy of 96.1% on AFW and 98.9% on FDDB. Then faces are sent to convolutional neural network for transfer learning because of limited training samples. On the testing dataset, the accuracy of face recognition reaches to 95.3%, comparing with 89.1% of the original network. What’ more, our method is better than the traditional algorithms and general CNN.