Content area
Full Text
Int J Comput Vis (2017) 122:426438 DOI 10.1007/s11263-016-0933-2
http://crossmark.crossref.org/dialog/?doi=10.1007/s11263-016-0933-2&domain=pdf
Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s11263-016-0933-2&domain=pdf
Web End = Deep Perceptual Mapping for Cross-Modal Face Recognition
M. Saquib Sarfraz1 Rainer Stiefelhagen1
http://orcid.org/0000-0002-1271-0005
Web End = Received: 15 January 2016 / Accepted: 6 July 2016 / Published online: 23 July 2016 Springer Science+Business Media New York 2016
Abstract Cross modal face matching between the thermal and visible spectrum is a much desired capability for nighttime surveillance and security applications. Due to a very large modality gap, thermal-to-visible face recognition is one of the most challenging face matching problem. In this paper, we present an approach to bridge this modality gap by a signicant margin. Our approach captures the highly non-linear relationship between the two modalities by using a deep neural network. Our model attempts to learn a nonlinear mapping from the visible to the thermal spectrum while preserving the identity information. We show substantive performance improvement on three difcult thermalvisible face datasets. The presented approach improves the state-ofthe-art by more than 10% on the UND-X1 dataset and by more than 1530% on the NVESD dataset in terms of Rank-1 identication. Our method bridges the drop in performance due to the modality gap by more than 40%.
Keywords Heterogeneous face recognition Visible face
recognition Night-time surveillance
1 Introduction
Face recognition, mainly, has been focused in the visible spectrum. This pertains to a large number of applications from biometrics, access control systems, social media tagging to person retrieval in multimedia. Among the main
Communicated by Xianghua Xie, Mark Jones and Gary Tam.
B M. Saquib Sarfraz
1 Institute of Anthropomatics & Robotics, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
challenges in visible face recognition, the different lighting/illumination condition has proven to be one of the big factors for appearance change and performance degradation. Many prior studies such as Li et al. (2007), Socolinsky and Selinger (2002), Nicolo and Schmid (2012), and Klare and Jain (2013) have stated better face recognition performance in the infra-red spectrum because it is invariant to ambient lighting. Relatively recently, few efforts have been devoted in the cross-modal face recognition scenarios, where the objective is to identify a person captured in infra-red spectrum based on its stored high resolution visible face image. The motivation for this...