{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/wasserstein-cnn-learning-invariant-features","title":"Wasserstein CNN: Learning Invariant Features for NIR-VIS Face Recognition","arxiv_id":"1708.02412","date":"2017-08-08","proceeding":null,"authors":["Ran He","Xiang Wu","Zhenan Sun","Tieniu Tan"],"abstract":"Heterogeneous face recognition (HFR) aims to match facial images acquired\nfrom different sensing modalities with mission-critical applications in\nforensics, security and commercial sectors. However, HFR is a much more\nchallenging problem than traditional face recognition because of large\nintra-class variations of heterogeneous face images and limited training\nsamples of cross-modality face image pairs. This paper proposes a novel\napproach namely Wasserstein CNN (convolutional neural networks, or WCNN for\nshort) to learn invariant features between near-infrared and visual face images\n(i.e. NIR-VIS face recognition). The low-level layers of WCNN are trained with\nwidely available face images in visual spectrum. The high-level layer is\ndivided into three parts, i.e., NIR layer, VIS layer and NIR-VIS shared layer.\nThe first two layers aims to learn modality-specific features and NIR-VIS\nshared layer is designed to learn modality-invariant feature subspace.\nWasserstein distance is introduced into NIR-VIS shared layer to measure the\ndissimilarity between heterogeneous feature distributions. So W-CNN learning\naims to achieve the minimization of Wasserstein distance between NIR\ndistribution and VIS distribution for invariant deep feature representation of\nheterogeneous face images. To avoid the over-fitting problem on small-scale\nheterogeneous face data, a correlation prior is introduced on the\nfully-connected layers of WCNN network to reduce parameter space. This prior is\nimplemented by a low-rank constraint in an end-to-end network. The joint\nformulation leads to an alternating minimization for deep feature\nrepresentation at training stage and an efficient computation for heterogeneous\ndata at testing stage. Extensive experiments on three challenging NIR-VIS face\nrecognition databases demonstrate the significant superiority of Wasserstein\nCNN over state-of-the-art methods.","url_abs":"http://arxiv.org/abs/1708.02412v1","url_pdf":"http://arxiv.org/pdf/1708.02412v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[],"tasks":[{"task_slug":"face-recognition","task_name":"Face Recognition"},{"task_slug":"heterogeneous-face-recognition","task_name":"Heterogeneous Face Recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/face-verification-on-buaa-visnir","task":"Face Verification","dataset":"BUAA-VisNir","model":"W-CNN He et al. (2018)","rank_in_archive_order":3,"of":3,"metrics":{"TAR @ FAR=0.001":"91.9","TAR @ FAR=0.01":"96.0"},"uses_additional_data":false},{"leaderboard":"/sota/face-verification-on-casia-nir-vis-20","task":"Face Verification","dataset":"CASIA NIR-VIS 2.0","model":"W-CNN He et al. (2018)","rank_in_archive_order":3,"of":3,"metrics":{"TAR @ FAR=0.001":"98.4"},"uses_additional_data":false},{"leaderboard":"/sota/face-verification-on-oulu-casia-nir-vis","task":"Face Verification","dataset":"Oulu-CASIA NIR-VIS","model":"W-CNN He et al. (2018)","rank_in_archive_order":3,"of":3,"metrics":{"TAR @ FAR=0.001":"54.6","TAR @ FAR=0.01":"81.5"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}