The process of establishing human identity based on the physical (e.g., fingerprints) or behavioral (e.g., gait) attributes of an individual constitutes the core of biometrics. Despite the successful deployment of biometric systems in several applications, a number of fundamental issues in biometrics are yet to be addressed in a comprehensive fashion. Indeed, establishing identity through biometrics is a complex and difficult problem due to the intrinsic challenges associated with the technology.
The objective of this research project is to systematically develop methods to overcome the limitations of existing biometric systems, thereby advancing the state of the art in this field. In this regard, the following tasks are being conducted: (a) modeling the biometric patterns of an individual using analysis-by-synthesis schemes, individuality analysis and manifold geometry; (b) designing efficient indexing methods for the rapid retrieval of identities from a large repository of biometric patterns; (c) enhancing the information content of biometric patterns using multibiometric fusion techniques; and (d) evaluating the performance of the aforementioned tasks on real-world biometric datasets. The models and techniques generated in this work are also applicable to other domains in pattern recognition and computer vision.
The results of this research project are expected to have a positive impact on the design and development of large-scale multibiometric systems for identity recognition management. In addition, the research activity will enhance the current biometrics curricula, engage students in cutting-edge research and promote use and effectiveness of biometric technology in diverse applications. The project's Web site (www.csee.wvu.edu/~ross/biometrics/) will be used for results and information dissemination to broad communities of researchers, educators, students and biometrics practitioners.
The outcomes of this project include the following: 1. Methods to perform biometric fusion: The use of multiple sources of biometric information (e.g., face and fingerprint, or multiple face matchers) to automatically recognize individuals is necessary to improve the performance of biometric systems operating in unconstrained environments. Biometric fusion is also important in the context of large-scale systems (e.g., national ID programs) where millions of individuals may be enrolled in the system. In this project we designed novel methods for (a) performing dynamic fusion based on quality of input biometric data; (b) combining biometric information at the rank level with applications in large-scale identification systems; (c) ranking the users enrolled in a biometric system based on error metrics, and utilizing this information to perform adaptive fusion; (d) incorporating the correlation structure between biometric classifiers when designing fusion rules; (e) predicting identification errors in a biometric system based on the evidence of match scores and ranks; and (f) incorporating liveness detection schemes with the biometric matcher in order to develop spoof-resistant systems. 2. Methods to perform biometric indexing: In a biometric identifcation system, the input biometric data has to be potentially compared with all the enrolled identities in the database in order to determine the best matching identity. This matching operation can be computationally prohibitive in systems dealing with millions of enrolled individuals. Indexing methods can be judiciously used to constrain the matching operation to a subset of the identities in the database. In this project, we designed novel methods for indexing large-scale fingerprint databases as well as large-scale multimodal biometric databases. Experimental analysis confirmed the potential of these methods in substantially reducing the number of matching operations that have to be conducted in the context of identification systems. We also designed classification and indexing methods for rapidly retrieving iris images from a large database. 3. Methods to impart privacy and security to biometric data: It is necessary to protect and secure the biometric data of an individual stored in a database. Towards this end, we designed a method based on Visual Cryptography, where the face image of an individual is decomposed into two component images in such a way that the original image can be revealed only when both component images are simultaneously available; further, each component image does not reveal the identity associated with the original image. This method ensures that the face image of an enrolled individual cannot be revealed unless there is authorization and cooperation between two independent parties hosting the biometric data. In addition, we designed a mixing scheme for securing and protecting fingerprint images. The proposed scheme mixes a fingerprint with another fingerprint (referred to as the "key") in order to generate a new mixed fingerprint image that can be directly used by a fingerprint matcher. The mixed image obscures the identity of the original fingerprint; further, different applications can employ different "keys", thereby ensuring that the identities enrolled in one application cannot be matched against the identities in another application. The concept of mixing was also used to generate joint digital identities, where a single fingerprint template was constructed from two separate fingerprints corresponding to two different identities. Joint digital identities can potentially be used to manage joint bank accounts or to perform group authentication. 4. Biometric education: The educational mission of this project resulted in the publication of a textbook titled "Introduction to Biometrics" by Jain, Ross and Nandakumar (Springer 2011). This textbook is currently being used by several instructors around the world in their classrooms. The book presents the fundamentals of a biometric system and dicusses the feature extraction and matching schemes of prominent biometric traits (e.g., fingerprint, face, and iris). The website associated with the textbook provides resources for students and instructors using this textbook.
