Adaptive Biometric Authentication using Dempster‐Shafer Networks: Concepts and Performance
Bojan Cukic, Natalia Schmid and Nick Bartlow (WVU)
Enhancing Iris Systems using Conjunctival Vascular Patterns
Reza Derakhshani (UMKC) and Arun Ross (WVU)
Fingerprint Matching Using Level 3 Features
Anil K. Jain (MSU)
Multispectral and Multiframe Iris Analysis: Phase II
Arun Ross, Lawrence Hornak and Xin Li (WVU)
Video‐based Metrology for Automated Human Identity Profiling
Don Adjeroh, Xin Li, Arun Ross and Bojan Cukic (WVU)
Sequential Testing for Biometric Error Rates
Michael Schuckers (St. Lawrence University)
Encryption of Biometric Templates using Biometrics as the Key
Stephanie Schuckers and S. Kumar (Clarkson)
Quality Assessment and Restoration of Face Images in Long Range/High Zoom Video
Besma Abidi (UK) and Natalia Schmid (WVU)
A Dynamic Hierarchical Fusion Architecture for Biometric Systems
Anil K. Jain (MSU) and Arun Ross (WVU)
Summaries:
Adaptive Biometric Authentication using Dempster‐Shafer Networks: Concepts and Performance
Bojan Cukic, Natalia Schmid and Nick Bartlow (WVU)
This project aims at providing a methodology for designing Dempster‐Shafer (D‐S) belief networks that optimize identity matching. D‐S belief networks can be automatically generated as system parameters and available evidence are updated. The strength of “knowledge” rests on a justification of belief acquired through the mathematical theory of evidence. Ideally the ability to automatically adapt will allow system performance to reach its full potential. This project will evaluate performance measures in these complex biometric systems, for example robustness and scalability, in order to understand whether the promise of performance improvement is justified. The measures will be compared against achievable limits.
Enhancing Iris Systems using Conjunctival Vascular Patterns
Reza Derakhshani (UMKC) and Arun Ross (WVU)
The conjunctival vascular structure of the eye will be used in conjunction with the iris pattern in order to validate the liveness of the iris, and enhance the recognition performance of an iris system.
Fingerprint Matching Using Level 3 Features
Anil K. Jain (MSU)
Fingerprint friction ridge details are generally described in a hierarchical order at three different levels, namely, Level 1 (pattern), Level 2 (minutiae points) and Level 3 (pores and ridge shape). Although Level 3 features are key for latent print examination and forensics research has shown the viability of using pores to assist identification, current Automated Fingerprint Identification Systems (AFIS) rely only on Level 1 and Level 2 features. With the advances in sensing technology, many commercial live‐scan devices are now equipped with high resolution (1000 ppi) scanning capability, allowing additional information besides minutiae, such as Level 3 features to be utilized. We propose a systematic study to determine how much performance gain one can achieve by automatically extracting and matching Level 3 features. Our initial experiments have shown that the use of Level 3 features provide a relative reduction of 20% in the equal error rate (EER).
Multispectral and Multiframe Iris Analysis: Phase II
Arun Ross, Lawrence Hornak and Xin Li (WVU)
Phase I of this project had explored the potential of using multispectral information to enhance the performance of iris recognition systems. In phase II the goal is to develop and implement algorithms that impart the following functionalities to an iris recognition system: iris localization using multispectral information; determining an optimal color space for iris texture analysis and processing; designing new algorithms for extracting novel features from various spectral channels; automatic clustering of iris components based on color information; fusing multispectral information based on eye color; and facilitating interoperability between iris images acquired at multiple wavelengths. By adopting this new research agenda, the PIs expect to advance the science and technology of multispectral iris analysis in the biometric domain.
Video‐based Metrology for Automated Human Identity Profiling
Don Adjeroh, Xin Li, Arun Ross and Bojan Cukic (WVU)
In this project, we propose to explore a new direction in human identification that does not rely on predefined patterns. That is, we construct a person’s biometric profile (BP) by extracting metrological information about the individual from a given observation data. Such a metrology‐based biometric profile is both dynamic (it varies along with the acquisition environment) and scalable (more measurements imply less uncertainty about the identity). The video metrology–based approach is particularly suitable for some applications where traditional biometrics is not applicable (e.g., masked or uncooperative individuals, or extreme environments, for instance, combat soldiers in full gear in a hazardous environment.
Sequential Testing for Biometric Error Rates
Michael Schuckers (St. Lawrence University)
In this project, we would like to develop and apply the statistical methodology for sequential testing to the testing of bio‐authentication devices. Specifically, we are interesting in testing whether or not a device’s error rate is below a given threshold. The basic idea of sequential testing, as it applies here, is that periodically testing will be stopped and evaluation of the error rate at that juncture will be determined. At each such stopping point, there are three options. First, if the error rate is sufficiently low, then the test is terminated and the device is said to meet the threshold. Second, if the error rate is sufficiently high, then the test is terminated and the device is said to not meet the threshold. Finally, if the error rate is neither sufficiently large nor extremely small then the test is continued until the next stopping point.
Encryption of Biometric Templates using Biometrics as the Key
Stephanie Schuckers and S. Kumar (Clarkson)
Biometric information is irrevocable and hence should not be compromised. With the advent of applications requiring transmission of biometric information using public networks for personal authentication, it has become necessary to embed strong security in the system. Previously, key‐based approaches have been suggested, but keys are often protected by passwords which often are chosen such that they are inherently weak. Our project studies methods to encrypt a biometric template by using biometric information itself instead of using keys. The proposed systems has these main components: Encoder: statistical template learning for a set of registered users to be used with partial biometric data to formulate encryption key; Simulation of a chaotic channel by mixing different biometric representations based on statistical properties of biometric data; Decoder: design of a blind source separator in tandem with hidden Markov model for fuzzy matching, trained to quantify the information related to the set of registered users; and an evaluation platform to assess system performance on real as well as manipulated database.
Quality Assessment and Restoration of Face Images in Long Range/High Zoom Video
Besma Abidi (UK) and Natalia Schmid (WVU)
The objective of this proposal is to design an efficient algorithm for the evaluation of face image quality in high magnification surveillance videos, then apply adaptive image deblurring and restoration to increase the quality of these face images so they become suitable for recognition. Tests will be conducted using FaceIt.
A Dynamic Hierarchical Fusion Architecture for Biometric Systems
Anil K. Jain (MSU) and Arun Ross (WVU)
The performance of a biometric recognition system can be significantly improved by combining multiple classifiers, by utilizing multiple samples during enrollment/authentication, or by including multiple biometric indicators. We propose to design a fusion framework that optimally combines information, possibly in a hierarchical way, pertaining to multiple samples and multiple classifiers (algorithms) in order to maximize the performance gain. To facilitate such a framework, we will investigate, in the context of face recognition, the diversity ( as well as quality) of information that is desired in the representative face samples (e.g., variations in pose, tilt, lighting, etc.), the nature of the face recognition algorithms to be combined (e.g., PCA, LFA, etc), and a dynamic hierarchical fusion architecture that determines the type of information to be fused as well as the fusion algorithm to be employed based on the available input data. We will also compare the performance gain of intra‐modal fusion (e.g., face alone) against inter‐modal fusion (e.g., face and fingerprint). This study will benefit dynamic surveillance applications and can be extended to include other modalities as well (such as fingerprints and iris).