Exploring the Science of Biometrics under Relaxed Constraints

The performance of biometrics algorithms is typically optimized by controlling, or constraining, as many acquisition parameters as possible while acquiring the desired features. It is usually well known which parameters have the greatest impact on performance, but very little scientific understanding of the phenomenology itself. As a result, it is often difficult to predict, and explain, how match algorithms will perform as these constraints are relaxed. This talk offer a perspective on the current state of the technology; and suggests areas of research to make biometric technologies more robust under relaxed constraints.