Advanced Methods and Applications for Fingervein Recognition

Fingervein recognition systems have been commercially deployed in several areas, in particular for financial services, as this biometric modality offers important advantages: Capturing is not sensitive to skin conditions and acquisition cannot be done without consent of the user. However, for these systems, not much is known about their claimed robustness towards varying acquisition conditions and the actual vessels recorded, in particular for transillumination-based systems. Also, an advantage often raised, i.e. the possibility of building contactless sensors, is not reflected by todays sensor designs. This project aims to develop a set of second generation sensors with corresponding biometrric recognition software to improve over todays systems. In particular, we will develop (i) two variants of contact-less sensors allowing to acquire five fingers concurrently and (ii) two variants of sensors allowing to capture fingervein related imagery in 3-D. With the latter data, we will develop techniques for reconstructing 3-D vessel structures and will develop corresponding techniques for feature extraction and template comparisons. The availability of 3-D sensing also changes the game in presentation attacks, aka sensor spoofing, as current techniques for conducting presentation attacks against these systems rely on 2-D printout vessel structures. We will (i) show that 3-D sensing cannot be fooled using this type of artifacts and (ii) develop 3-D presentation attack artifacts to test if these can be used to fool the developed 3-D sensors. Finally, we will establish presentation attack detection techniques based on detecting blood flow in NIR-imagery (as opposed to using global motion as done in current techniques), which will also be able to detect the developed 3-D artifacts. The project will also investigate if fingervein recognition can be used in a forensic post-mortem victim identification scenario. For this purpose, the devloped sensors will be used to acquire post-mortem imagery from corpses in increasing time-steps after death to investigate the feasibility of this approach. Besides using biometric sensors, classical medical image acquisition will be used as a baseline (i.e. magnetic resonance angiography MRA) and techniques will be developed to conduct biometric comparisons among the different sensor types.

Fingervein recognition systems have been commercially deployed in several areas, in particular for financial services, as this biometric modality offers important advantages: Capturing is not sensitive to skin conditions and acquisition cannot be done without consent of the user. However, for these systems, not much is known about their claimed robustness towards varying acquisition conditions and the actual vessels recorded, in particular for transillumination-based systems. Also, an advantage often raised, i.e. the possibility of building contactless sensors, is not reflected by todays' sensor designs. This project aims to develop a set of second generation sensors with corresponding biometrric recognition software to improve over todays' systems. In particular, we will develop (i) two variants of contact-less sensors allowing to acquire five fingers concurrently and (ii) two variants of sensors allowing to capture fingervein related imagery in 3-D. With the latter data, we will develop techniques for rotation-invariant recognition and will develop corresponding techniques for feature extraction and template comparisons. The availability of rotation-invariant sensing also changes the game in presentation attacks, aka sensor spoofing, as current techniques for conducting presentation attacks against these systems rely on 2-D printout vessel structures. We will (i) show that our sensing cannot be fooled using this type of artifacts and (ii) develop 3-D presentation attack artifacts to test if these can be used to fool the developed 3-D sensors. Finally, we will establish presentation attack detection techniques based on detecting blood flow in NIR-imagery (as opposed to using global motion as done in current techniques), which will also be able to detect the developed 3-D artifacts. The project will also investigate if fingervein recognition can be used in a forensic post-mortem victim identification scenario. For this purpose, the devloped sensors will be used to acquire post-mortem imagery from corpses in increasing time-steps after death to investigate the feasibility of this approach. Besides using biometric sensors, classical medical image acquisition will be used as a baseline (i.e. magnetic resonance angiography MRA) and techniques will be developed to conduct biometric comparisons among the different sensor types. An important topic is the development of techniques to improve security and privacy, i.e. template protection and sample encryption techniques, which are particularly motivated by discussed template-inversion and morphing attacks, respectively.