Developing a context-dependent tuning framework of multi-channel biometrics that combine audio-visual characteristics for secure access of an ehealth platform

Abstract

The efficiency of a biometric system is identified by the detection error tradeoff (DET) curve, which is a visual characterization of the trade-off between the False Acceptance Rate (FAR) and the False Rejection Rate (FRR). A DET curve is a plot of FAR against FRR for various threshold values, t. FRR refers to the expected probability that two mate samples (samples of the same biometric trait obtained from the same user) will be falsely declared as a non-match whereas FAR is the expected probability that two non-mate samples will be incorrectly recognized as a match. The threshold t defines how much the biometric characteristics must be similar, in order to make a positive comparison, so it measures the correspondence between characteristic to check and template stored in the database. By elevating the threshold, the risk that not authorized users can fool the system diminishes, but, on the other hand, it is more probable that some authorized users can sometimes be refused. In this work, we present the results for SpeechXRays multi-modal biometric system that uses audio-visual characteristics for user authentication in an eHealth platform for osteoarthritis management. Using the privacy and security mechanism provided by SpeechXrays based on audio and video biometrics medical personnel is able to be verified and subsequently identified to the eHealth application for osteoarthritis.