Multi-task and Multi-scale Face Recognition Based on CNN

Face recognition has been widely used in social security, privacy protection and biometrics due to its advantages. However, under wild world condition face recognition task can be very complex due to illumination variation, occlusion, facial expression, etc. Considering these, this paper proposes a multi-scale feature fusion convolutional neural network combined with multi-task learning. The face recognition main task is decomposed into pose estimation, illumination classification and occlusion classification subtasks, which are jointly used to promote the optimization of the main task. Then, the weight distribution strategy of the loss function among subtasks and its influence are discussed. Experiment results show that the proposed method achieves good performance on the public face verification dataset and our own face verification dataset.