10/12/2006
3:30pm-4:30pm
ECCR 265
|
Improving Biometric System Privacy, Security and Accuracy
University of Colorado at Colorado Springs
The talk will start with a motivation for the need for improved
Security/Privacy in Biometrics. This section of the talk will briefly review
the privacy/security issues with biometric, explain why standard encryption
does not solve the problems and quickly review the state of the art in privacy
preserving biometrics. It will then present the key ideas behind the
patent-pending Biotopes®, and present performance results where existing
face and fingerprint algorithms were extended with Biotopes to improve
security/privacy. We discuss why the Biotopes and associated transforms,
actually improved the accuracy of the underlying algorithms. We conclude this
segment on privacy/security with some examples highlighting the important of
accuracy and why it has such a strong impact on privacy/security concerns.
The second aspect of the presentation will introduce our unique work looking at
predicting the failure of a biometric recognition system. The approach, Feature
Analysis of Similarity Surface Theory (FASST), conjectures that the the
similarity scores used in recognition contain information which can, in general,
predict when the system is failing. AdaBoost is used to combine the features
computed from the similarity surface to produce a patent pending system that
predicts the failure of a biometric system. Face-system Failure prediction,
using a leading leading commercial face recognition system, is presented as an
example to show how to use the approach. On outdoor weathered face data, the
system demonstrated the ability to predict 90% of the underlying facial
recognition system failures with only a 15% false alarm rate.
The final component of the talk presents our Random-Eyes® approach, a
novel technique for improving face recognition performance by predicting
system failure, and, if necessary, perturbing eye coordinate inputs and
re-predicting failure as a means of selecting the a perturbation that
provides correct classification. This relies on a method that can accurately
identify patterns that can lead to more accurate classification, without
modifying the classification algorithm itself. Showing the generality of
FASST, this time we use a neural network trained on wavelet transforms of
similarity score distributions from an analysis of the gallery. Face images
with a high likelihood of having been incorrectly matched are reprocessed
using perturbed eye coordinate inputs, and the best results used to
"correct" the initial results. Results for both commercial and research
face-based biometrics are presented using both simulated and real data. The
statistically significant results show the strong potential for this to
improve system performance, especially with uncooperative subjects.
The talk will end with a brief five-minute overview of the VAST Lab at UCCS and
our many other projects ranging from video surveillance networks, to
distributed steganalysis, to low-power high-bandwidth sensor networks,
to embedded systems.
Hosted by Jane Mulligan.
|
