Multi-feature deep learning for face gender recognition

Face gender recognition is a challenging problem in the traditional field of pattern recognition. In this paper, we propose a deep learning model that can learn the joint high-level and low-level features of human face to address this problem. Our deep neural networks apply convolution and subsampling in extracting the local and abstract features of human face, and reconstruct the raw input images to learn global and effective features as supplementary information at the same time. We also add a trainable weight in the networks when combining the two kinds of features to realize the final gender classification. Experiment results show that our method achieves the highest accuracy compared with existing methods, when test on the mixed face dataset. Further, in the generalization test, the average classification rate on 3 public datasets of our method is 5% higher than the joint Local Binary Pattern (LBP) and Support Vector Machine (SVM) method, and is nearly 1% higher than the SVM with face pixels method. This proves our method outperforms the traditional methods in both learning ability and generalization ability.