Relational Deep Feature Learning for Heterogeneous Face Recognition

Heterogeneous Face Recognition (HFR) is a task that matches faces across two different domains such as visible light (VIS), near-infrared (NIR), or the sketch domain. Due to lack of databases, HFR methods usually exploit pre-trained features on large-scale visual database contain general facial information. However, these cause performance degradation due texture discrepancy with With this motivation, we propose graph-structured module called Relational Graph Module (RGM) extracts global relational information in addition features. Because each identity's between intra-facial parts similar any modality, modeling relationship can help cross-domain matching. Through RGM, relation propagation diminishes dependency without losing its advantages from Furthermore, RGM captures geometrics locally correlated convolutional identify long-range relationships. In addition, Node Attention Unit (NAU) performs node-wise recalibration concentrate more informative nodes arising relation-based propagation. suggest novel conditional-margin loss function (C-softmax) for efficient projection learning embedding vector HFR. The proposed method outperforms other state-of-the-art five databases. demonstrate improvement three backbones because our be plugged into face recognition backbone overcome limitations small database.