Obtaining ROC Curves from
Available Biometrics Systems



This project requires no programming. Once you understand the idea of Receiver Operating Characteristic (ROC) curves, which will be made clear by your customer, you will find this to be an interesting and enjoyable project.

There are several biometric system demos freely available for public use. Some of these biometric systems provide a classifier output score, which is the strength of the match with the enrolled template. Although many of these systems do not report error rates (i.e. they avoid actually making an authentication decision), the classifier scores can be used to determine system performance.

Finding biometric systems to test can be a challenge. Some commercial systems are only available upon request. Other systems are from smaller research groups and may not be indexed by Google. You may have to reach out to other research groups and commercial biometric companies to request a demo. Start with keystroke and mouse biometric demos, then move on to other biometric systems per the customer's recommendations.

To evaluate a biometric system, it is suggested to obtain authentic and impostor scores from at least 10 people. First, obtain scores from within your team. Then work with other teams to gather more scores. To help centralize the score collection, a public Google doc can be used to record scores (e.g. you might create one doc for the authentic scores and one for the impostor scores and assign a row for each person to edit).

Behavioral privacy is also becoming a concern as more behavioral biometric systems are being deployed. A browser plugin for Chrome (link below) was recently developed to try to mitigate the ability to record typing behavior. System performance for the available keystroke biometric demos should be obtained using the plugin with various settings to determine whether it really works.