Multi-Scale Thermal to Visible Face Verification via Attribute Guided Synthesis

Thermal-to-visible face verification is a challenging problem due to the large domain discrepancy between modalities. Existing approaches either attempt synthesize visible faces from thermal or learn domain-invariant robust features these modalities for cross-modal matching. In this paper, we use attributes extracted images attribute-preserved imagery A pre-trained attribute predictor network used extract image. Then, novel multi-scale generator proposed image guided by attributes. Finally, VGG-Face leveraged synthesized and input verification. Extensive experiments evaluated on three datasets (ARL Face Database, Visible Thermal Paired Tufts Database) demonstrate that method achieves state-of-the-art performance. particular, it around 2.41\%, 2.85\% 1.77\% improvements in Equal Error Rate (EER) over methods ARL respectively. An extended dataset Dataset volume III) consisting of polarimetric 121 subjects also introduced paper. Furthermore, an ablation study conducted effectiveness different modules method.