Deep Learning vs Hand-Crafted Features for Face Recognition under Uncontrolled Conditions

Facial recognition in degraded conditions has become a very active and widely used area in several fields such as video surveillance and access control. Indeed, the presence of facial degradation such as the variation in face expressions, head poses, illumination or presence of partial occlusion, presents a great challenge for facial recognition. To overcome this problem, several approaches and techniques have been proposed in the literature. In this paper, we make a comparison between usual hand-crafted features, pre-trained architecture models and learned features extracted from the final layers of these pre-trained models. In particular, exploiting the large learning capability of deep networks improves the facial recognition results compared to usual methods based on handcrafted features. The results based on the Eurecom Kinect Face DB dataset demonstrate the effectiveness of learned features for face recognition classification.