Temporal Integration for Continuous Multimodal Biometrics


Temporal Integration for Continuous Multimodal Biometrics

Alphan Altinok, Matthew Turk

Continuous Multimodal Biometric Authentication.


Continuous multimodal biometric authentication aims to report a degree of belief in the user's identity based on multimodal biometric data acquired over time.


We present an approach for temporal integration based on uncertainty propagation over time for estimating channel output distribution from recent history, and classification with uncertainty. Our method operates continuously by computing expected values as a function of time differences. Scores (belief measures in the user's identity) from individual biometric channels are treated as Gaussians centered at the value of each score, which are propagated in time with increasing uncertainty as shown in the figure.

Temporal integration is based on uncertainty propagation. The degree of belief in the validity of a score decreases (uncertainty about the score increases) over time according to a predefined degeneracy function. This formulation enables us to integrate multimodal data by classifying them with respective uncertainties regardless of the observation time. In Figure 2, vertical lines show the variances of propagated distributions around the means since the last score.


A. Altinok, M. Turk.
Temporal Integration for Continuous Multimodal Biometrics.
In Multimodal User Authentication '03, Santa Barbara, CA, Dec. 11-12, 2003.