Next generation biometrics : line scan palmprint recognition and door knob hand recognition

PolyU Library Call No.: [THS] LG51 .H577P COMP 2016 Quxiii, 58 pages :color illustrationsBiometrics has been in the spotlight for over a decade. A biometric system authenticates people's identities using biological and/or behavioral traits. Various biometric applications have been developed, and a vast amount of people have bene.ted from the convenience of biometrics. The applications of biometrics in everyday life have expanded dramatically. In the meantime, there are new challenges to biometric systems when being used by the general public. In this work, the focus is to investigate the methodology of conventional biometrics design, to improve current biometrics, to practice new biometrics design methodology, and to create new biometrics. Palmprint recognition is the foundation of this improvement. Palmprint recognition is a hand-based biometrics with excellent performance for the large region of interest, stable and unique features, and fast feature extraction and matching method verified by researchers all over the world. However, the applications of palmprint recognition are limited. One major reason for the palmprint recognition not being utilized widely is the lack of a small and compact acquisition system. Then we proposed a line-scan palmprint acquisition system (LPS). LPS uses line scan imaging that makes the system small and flexible. LPS comprises the line scan sensor, the synchronizing unit, and the controller board. With the proposed acquisition system, we collected 8000 images from 250 people. In this huge data set, the equal error rate of LPS is 0.048%, which is comparable to conventional palm-print acquisition systems. Moreover, the volume of the proposed LPS is 94% smaller than conventional systems at least. After the line scan palmprint system, it is realized that a new biometrics design model and new biometric systems are required. This new biometrics design model should emphasize user experience, because the improvements in system level is not enough. The conventional biometrics design model focuses on security performance. They are under the assumption that biometric traits can be extracted better in the standard pose and position than in a deformed pose. This assumption is true with the limitations of both hardware (imaging technology) and algorithms (feature extraction and classi.cation methods). With the recent development in both hardware and algorithms field, this situation has altered. For hardware, the lens and cameras have become tiny in volume. The precision and the resolution of the lens and cameras are far better than five years ago. For algorithms, new feature extraction method and classification method are robust to various interference and deformations. Thus, we proposed the new ergonomic biometrics design model. In this ergonomic biometrics design model, the human factor, both physical and psychological factors, is one of the major concerns in the design stage of building a biometric system. Using this model, we proposed the door knob hand recognition system (DKHRS).The acquisition device of DKHRS has an identical appearance of a conventional door knob. It capturesa hand image when people hold the door knob. The door knob hand recognition is user-friendly in both physical and psychological merits. First, the appearance of door knobs has been evolved for over a hundred years. The design of door knob is one of the most convenient design of a device. Second, the function of door knobs is access control. This function and interactions of a door knob have been accepted by the human society for over a century. Following this convention, people could understand the function of the system in a very short time. Third, using biometric feature instead of conventional keys and locks empowers people the convenience of flawless security experience, because the biometric feature can not be lost or forgotten. In DKHRS, there is a critical problem: imaging. The imaging is a challenge because the system requires an omnivision image capturing in a very compact space. Thus, we proposed a simplified catadioptric imaging scheme, which can capture omnivision images but is also easy to manufacture and cost-efficient. Using this imaging scheme, the door knob hand recognition acquisition device is established. It comprises six components: an acrylic transparent door knob, an over 95% full spectrum reflective mirror, a white LED board, a 6 mm focal length pinhole lens, a 1/3 inch mini CCD camera, and a USB 2.0 frame grabber. With the developed acquisition system, a big door knob hand image database containing 10920 samples from 420 hands is built. The feature extraction and classification method is another major challenge. After investigating state-of-art feature extraction methods for hand-based biometrics, we propose a patch-based method which combines the local Gabor binary pattern, the principal component analysis, and the projective dictionary pair learning method. In a large data set including 420 hands and 10920 images, the proposed methods achieves an 0.091% equal error rate. The performance is promising and competitive in biometric applications.Department of ComputingPh.D., Department of Computing, The Hong Kong Polytechnic University, 2016Doctorat