FASST Multimodal Biometric Screening

Multimodal biometrics for screening presents unique challenges. The evaluation of them presents even more challenges. The Securics team is leveraging 7 years of effort and past DARPA HID funding, on unique biometric evaluation approaches for measuring and comparing failure/performance predictions. The approach models what is important for a particular task and then learning the coefficients of the reliability measures from empirical datasets. It extends our past work on Failure Analysis from Similarity Surface Theory. The learning/training allows the measures to adapt to population, sensor, and algorithm specific properties. The resulting algorithm can be applied in real time. The effort includes two unique approaches to experimentation for large-scale multi-modal evaluation of non-cooperative and adversarial subjects. A second important and unique, aspect of our work has been the development of formal metrics for comparison of biometric reliability measures and fusion. While multiple papers develop quality-based measures and fusion techniques, there has been no paradigm or framework for quantitative comparison. This Phase I SBIR presents a unique opportunity to advance the state of the art both in multi-modal biometric data quality by developing a toolset in which different reliability can be quantitatively combined and compared on real data from real algorithms.