Multi-Modal Biometric Verification
Security of Cyber-Physical Systems
Multi-Modal Biometric Verification
Motivation: Verifying the claimed identity of an individual is critical in providing secure access to physical systems, devices and spaces. Most current verification systems are password based making them susceptible to problems such as forgetting the password and passwords being stolen. One way to overcome these problems is to employ biometrics (e.g., fingerprints, face images, iris patterns, etc.) for authentication. However, verification methods based on a single biometric type or algorithm may not provide adequate verification performance, can be subject to attacks and not every one may be able to use a particular biometric. For example, some face recognition algorithms cannot handle expressions whereas other algorithms may have problems with illumination changes. We can improve the overall performance by employing different algorithms and intelligently fusing the results. Also, in some cases, one biometric may not be applicable at all. For example, some people don’t have good enough ridges in their fingerprints to be able to use fingerprint verification systems. We believe that by using multiple biometric modalities, we will achieve not only improved security, but also a verification system that can be used by more people. The goal of the proposed research effort is to develop, evaluate and demonstrate methods for multi-modal biometric verification.
Accomplishments in 2003: There are three main biometric modalities that produce images and thus allow the application of advanced image processing algorithms. Last year, we started the multi-modal biometric verification project with C3S support. As face and fingerprint verification have received significantly more research attention than iris verification, one of the goals of this project is to develop and evaluate iris verification methods. Another goal is to investigate methods to use multiple biometrics to achieve significantly better verification performance. Following were accomplished during 2003 towards these goals.
- To develop and evaluate iris verification algorithms, we need an iris database as no such database is publicly available. We have identified, acquired and set up a camera system in the biometrics lab in Hamburg Hall to acquire iris images. We have applied for IRB approval for this data collection process as human subjects are involved. Once the approval is received, the plan is to collect multiple images of the two irises of about 50 subjects. We plan to make this database available for use by other research groups to foster independent evaluation.
- We have investigated the use of correlation filters for iris verification. Results on a somewhat small iris database are encouraging and we have presented these at the Audio- and Video-based Person Authentication (AVBPA) conference held in July 2003.
- For multi-modal biometric verification, it is convenient to use a common solution methodology for multiple biometrics. Towards that goal, we have investigated the use of correlation filters for fingerprint verification and presented these results at the Multi-modal User Authentication (MMUA) workshop held in December 2003.
- We have extended our research in face recognition in many ways including reducing the complexity, making the templates cancelable, making the recognition systems illumination-tolerant, using video sequences, using multiple cameras and using color. We presented these results at various conferences. See the attached list of our papers.
- We have started developing algorithms for multi-biometric fusion. One of our early observations is that it is better if the verification errors from different algorithms and/or different biometric types are negatively correlated than if they are independent.
- We presented 4 papers at the 2003 Audio- and Video-Based Person Authentication (AVBPA) conference in England, 1 paper at the International Conference on MultiMedia (ICME), and 2 papers at the Multi-modal User Authentication (MMUA) workshop.
- We have 4 papers accepted for presentation at the first SPIE conference on Biometrics Technology for Human Identification and 2 papers accepted for the first International Conference in Biometric Authentication (ICBA).
- We have submitted 3 papers to Intl. Conf. on Image Processing (ICIP), 2 papers to Computer Vision and pattern Recognition (CVPR) and 2 to International Conference on Pattern Recognition (ICPR).
Plans for 2004: The goals for this project during 2004 are as follows.
- Complete the acquisition of iris database and start making it available for other biometrics researchers,
- Improve the iris verification algorithms,
- Comparative analysis of methods for fusing multiple biometric types,
- Compare different methods for fusing multiple algorithms for one biometric type, and
- A demo that illustrates the benefits of multi-modal biometric verification system.