Biometric Image / Signal Processing and Pattern Recognition

Funded Projects

Participants

Keywords

multimodal biometrics, spoof, liveness, automated dental ID, fingerprint matching, biomedical signal ID, questioned documents

Area Description

Our proposed research will address the preprocessing (e.g., image enhancement), feature extraction (e.g., learning-based technique for fingerprint minutia extraction) and matching (e.g., using multimodal techniques for identification) functions of biometric systems. We also propose to address four rather difficult yet important applications of biometrics including biomedical systems.

Multimodal Biometrics
Every biometric indicator has some limitations. For example, it is estimated that approximately 5% of the population does not have "legible" fingerprints. Current face recognition systems are not robust to changes in ambient illuminations and pose of the subject. A multimodal system, which combines the decisions made by a number of independent biometrics indicators can overcome some of these limitations. A multimodal system is generally more robust to fraudulent technologies, because it is more difficult to forge multiple biometric characteristics than to forge a single biometric characteristic. Research issues include developing algorithms for combining multiple biometrics to address different types of applications ranging from low-security, high volume application such as access to ATM machines to high-security, low volume application such as access to a missile site. In addition, for high security applications, the biometric system may have the need for a "liveness" test for detection of "spoof attacks". We will analyze biometric features for life signs in addition to standard medical vital signs measurements.

Minutia Verification and Classification for Fingerprint Matching
Most of the existing automatic fingerprint verification systems first detect the minutiae in a fingerprint image and then match the input minutias set with the stored template. For poor quality images, almost all of the available algorithms detect a large number of false minutiae, leading to poor matching performance. We propose a feedback system for minutiae extraction which is based on an analysis of the gray scale profile in the neighborhood of potential minutiae. We also propose a feature refinement stage where the minutiae are classified into two major classes: ridge bifurcation and ridge ending. The goal of the feedback system is to learn the characteristics of minutiae in gray level images which can then be used to verify each detected minutia. This step will replace the rather ad-hoc minutia-pruning stage commonly used in fingerprint matching systems. Our matching algorithm will also be modified to match minutiae of the same type in the sensed image and the template. These two modifications in the feature extraction stage are likely to improve the fingerprint matching performance.

Latent Fingerprint Image Identification
The objective of this research is to develop a methodology for latent (partial) fingerprint image identification. This is a common scenario when the fingerprint images are lifted from a crime scene. Our work will involve developing a new and efficient algorithm when the number of common minutiae points between the sensed image and the stored templates is small (e.g., ~10). The approach for solving the latent fingerprint identification will also be useful when the sensed images are heavily distorted (e.g., due to finger cuts)or the acquired images have poor contrast (due to dry fingers).

Automated Dental Identification System (ADIS)
The objective of this project is to develop a prototype and a test bed for an ADIS system. The ADIS system will be used to identify missing persons using a database of dental scans. The system will compare a test denture scan image with a database of denture scans of missing persons. The challenge here is image enhancement and feature extraction. This research will be conducted in collaboration with Dr. Robert Howell, School of Dentistry at West Virginia University. This project was initiated based on FBI input.

Questioned Document Analysis
This application deals with determining the authorship of written documents. Here the research will focus on developing techniques to recognize (match) writer-dependent features in document images. Our system should help in detecting forgeries and identifying the culprit.

Biosignal Identification for Medical ID Applications
In the medical application domain, automated identification systems are central to systems requiring automated diagnosis. Signal processing is the basis for developing ways to simplify complex physiologic signals for diagnostic purposes. In addition, long-term biomedical signals, for many months or years for research or telemedicine applications, need further processing to integrate a vast amount of information for ease of human perusal and understanding. One example of research in this area is the development of automated algorithms for detection of arrhythmias, abnormal cardiac electrical abnormalities that cause a heart attack[6]. These automated algorithms are used in "smart medical devices" which make therapeutic decisions automatically and actually deliver the therapy. Medical applications are numerous as technology infuses the medical community with advances in communication, computational power, and miniaturization. Specialized techniques for biomedical signals need to be developed in the areas of signal recovery, feature extraction, algorithm creation, classification, testing, analysis of large temporal data sets, and data mining.