Skip to main content

Multi-Biometric Authentication System (MBAS)

Researcher: Vijayakumar Bhagavatula


Multi-Biometric Authentication System (MBAS)

Verifying the identity of a user is critical for many security applications such as e-commerce, access control and surveillance. Most current authentication 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, iris, voice, etc.) for authentication. A major challenge in using biometrics is their normal variability.

For example, a person's face can appear differently under different illuminations, expressions, and orientations. A useful biometric verification system must provide acceptable tradeoffs between false non-match rates due to such normal variability and false match rates due to impostors. One way to improve the verification performance is to use more than one biometric. Thus, this project is aimed at developing a robust authentication system using multiple biometrics.

Although the focus of this project is on verification (also known as one-to-one matching), much of the technology being developed is very applicable to biometric identification (also known as one-to-N matching). Recently, we have been able to demonstrate equal error rates (this refers to the setting where the false nonmatch rates and false match rates are equal) of about 0.15% on a face database (collected at the Advanced MultiMedia Processing Lab at CMU) of 13 individuals, each with 75 images of varying expressions. Only three images from each person were used for the enrollment. We have also achieved zero verification errors on the illumination subset of PIE (pose, illumination and expression) database of faces collected at CMU.

We are obtaining similar excellent verification results for fingerprint verification. Our near term plans include testing our technologies for biometric identification and applying these to voice recognition and iris recognition. Finally, we are also developing methods to reduce the enrollment and verification complexity of our methods so that they can be accommodated in a variety of systems.

Sony, General Motors, NIST