Date of Graduation
Statler College of Engineering and Mineral Resources
Lane Department of Computer Science and Electrical Engineering
Face biometrics leverages tools and technology in order to automate the identification of individuals. In most cases, biometric face recognition (FR) can be used for forensic purposes, but there remains the issue related to the integration of technology into the legal system of the court. The biggest challenge with the acceptance of the face as a modality used in court is the reliability of such systems under varying pose, illumination and expression, which has been an active and widely explored area of research over the last few decades (e.g. same-spectrum or homogeneous matching). The heterogeneous FR problem, which deals with matching face images from different sensors, should be examined for the benefit of military and law enforcement applications as well. In this work we are concerned primarily with visible band images (380-750 nm) and the infrared (IR) spectrum, which has become an area of growing interest.;For homogeneous FR systems, we formulate and develop an efficient, semi-automated, direct matching-based FR framework, that is designed to operate efficiently when face data is captured using either visible or passive IR sensors. Thus, it can be applied in both daytime and nighttime environments. First, input face images are geometrically normalized using our pre-processing pipeline prior to feature-extraction. Then, face-based features including wrinkles, veins, as well as edges of facial characteristics, are detected and extracted for each operational band (visible, MWIR, and LWIR). Finally, global and local face-based matching is applied, before fusion is performed at the score level. Although this proposed matcher performs well when same-spectrum FR is performed, regardless of spectrum, a challenge exists when cross-spectral FR matching is performed. The second framework is for the heterogeneous FR problem, and deals with the issue of bridging the gap across the visible and passive infrared (MWIR and LWIR) spectrums. Specifically, we investigate the benefits and limitations of using synthesized visible face images from thermal and vice versa, in cross-spectral face recognition systems when utilizing canonical correlation analysis (CCA) and locally linear embedding (LLE), a manifold learning technique for dimensionality reduction. Finally, by conducting an extensive experimental study we establish that the combination of the proposed synthesis and demographic filtering scheme increases system performance in terms of rank-1 identification rate.
Osia, Nnamdi C., "Unifying the Visible and Passive Infrared Bands: Homogeneous and Heterogeneous Multi-Spectral Face Recognition" (2017). Graduate Theses, Dissertations, and Problem Reports. 6358.