Tech Talk #3: Stephen Elliott (Summary)
Associate Professor Stephen Elliott, Industrial Technology, Purdue University
Director, Biometric Standards, Performance and Assurance Laboratory
Summary by Kelley Misata
Title: Advances in Biometric Testing
Starting the conversation Stephen reminded the audience that what makes biometrics such an interesting field is the unpredictability of the humans in the testing and evaluations processes. In traditional biometric testing environments researchers work with algorithms and established metrics and methodologies. However, as biometrics testing moves to operational environments there are more uncertainties to content with and therefore making it hard to do. Considering these two important testing environments, what biometric researchers are now trying to do is to understand further how a biometric system performs in any environment and identify what (or who) could the possible cause of errors.
As Stephen pointed out, there have been several papers addressing how individual error impacts biometric performance and the potential causes of these errors. Some of these errors are now being traced to gaps in biometrics testing including training (e.g. "How do you train someone who is difficult to train or doesn’t want to be trained?"), accessibility (e.g. "Are the performance results different in a operation environment than collected in a lab?"), usability (e.g. "Can the system be used efficiently, effectively and consistently by a large population?") and the complexities of the human factors on biometric testing performance. Raising the question, is the error always subject centric?>
In order to fill in some these gaps, Stephen and his graduate students are looking at the traditional biometric modes and metrics to determine if they are suitable in today’s testing and evaluation environments. During the CERIAS tech talk Stephen spotlighted the research of three of his graduate students: 1. The Concept of Stability Thesis by Kevin O’Connor - the examination of finger print stability across force levels; 2. The Case of Habituation by Jacob Hasselgren - quantitatively measuring habituation in biometrics testing environments; and 3. Human Biometric Sensor Interaction highlighting Michael Brokly’s research on test administrator errors in biometrics, including the effects of operator train, workloads of both test administrators and test operators, fatigue and stress.
The biometrics community continues to investigate these questions in order to understand how the vast array of players in a operational data collection environment impact performance. In his closing statements, Stephen reiterated the complexities and challenges in biometrics testing and how researchers are looking deeper into the factors affecting performance beyond a simple ROC/DET curve.