2020/04/30 by Xing Di, Benjamin S. Riggan, Shuowen Hu +2 · 37 citations
Computer Science · #Face (sociological concept) #Face and Expression Recognition #Face recognition and analysis #Facial recognition system #Generative Adversarial Networks and Image Synthesis #Generator (circuit theory) #Image (mathematics) #Pattern recognition (psychology) #Thermal #Visible spectrum #cs.CV
paper · pdf · doi:10.1109/tbiom.2021.3060641
published in IEEE Transactions on Biometrics Behavior and Identity Science 3(2), 266-280 (Institute of Electrical and Electronics Engineers) · accepted by IEEE Transactions on Biometrics, Behavior, and Identity Science (T-BIOM). arXiv admin note: substantial text overlap with arXiv:1901.00889
openalex created_date 2020/05/01 · arxiv created 2021/02/14 · openalex publication_date 2021/02/18 · arxiv updated 2021/04/12 · openalex updated_date 2026/07/28
Thermal-to-visible face verification is a challenging problem due to the large domain discrepancy between the modalities. Existing approaches either attempt to synthesize visible faces from thermal faces or learn domain-invariant robust features from these modalities for cross-modal matching. In this paper, we use attributes extracted from visible images to synthesize attribute-preserved visible images from thermal imagery for cross-modal matching. A pre-trained attribute predictor network is used to extract the attributes from the visible image. Then, a novel multi-scale generator is proposed to synthesize the visible image from the thermal image guided by the extracted attributes. Finally, a pre-trained VGG-Face network is leveraged to extract features from the synthesized image and the input visible image for verification. Extensive experiments evaluated on three datasets (ARL Face Database, Visible and Thermal Paired Face Database, and Tufts Face Database) demonstrate that the proposed method achieves state-of-the-art performance. In particular, it achieve around 2.41%, 2.85% and 1.77% improvements in Equal Error Rate (EER) over the state-of-the-art methods on the ARL Face Database, Visible and Thermal Paired Face Database, and Tufts Face Database, respectively. An extended dataset (ARL Face Dataset volume III) consisting of polarimetric thermal faces of 121 subjects is also introduced in this paper. Furthermore, an ablation study is conducted to demonstrate the effectiveness of different modules in the proposed method.