Face Recognition Method Based on Deep-shallow Feature Fusion

Aiming at the problem of limited discrimination of features extracted by traditional shallow features and rapid decline of recognition rate in complex environments, a face recognition algorithm that fuses the deep and shallow features is proposed in this paper. We extract the HOG feature and PCANet feature of the face image respectively, and used Principal Component Analysis (PCA) method to reduce the dimensionality of the feature vectors. After fusing the processed feature vectors, we used Linear Discriminant Analysis (LDA) to further reduce the dimensionality and select the discriminative information. And SVM classifier is adopted for classification. The experimental results on AR and LFW dataset show that our algorithm has higher recognition rate and stronger robustness to complex environments than the single feature method.