MASKS: Maintaining Anonymity by Sequestering Key Statistics

Customized Human Face Masking for Better Security and Privacy

Despite the explosion in the application of biometric technologies, there seems to have been surprisingly little serious attention paid to ensuring the security and privacy of these applications, whether as a matter of policy or technological research. Like with most technology, particularly technology available on the open market, there is no way to make its user have benevolent intentions. It is not hard to imagine how the technology could be misused with devastating effect against, for example, people enrolled in the Federal Witness Protection Program, or operatives of covert intelligence agencies.

The MASKS project was created with a goal of establishing some basic principles for the secure handling of biometric data. Our methods for finding such principles are deliberately adversarial, a so-called red team approach. We are trying to understand how to reliably prevent a biometric recognition system from performing as intended. The knowledge gained in such an approach can then be used to thwart deliberate misuse of biometric data, as well as allowing us to improve the reliability of biometric components of our own secure systems.

While the project is currently focused specifically on studying human facial images and the automated systems for recognizing them, we are developing theoretical principles general enough to encompass other biometrics, and possibly other personally identifiable patterns as well, such as shopping habits.

In addition to the information on this page, the published accomplishments of the project can be reviewed in the following papers:

• James Alexander and Jonathan Smith, Engineering Privacy in Public: Confounding Face Recognition, In The Third Workshop on Privacy Enhancing Technologies, 2003.
• James Alexander, The MASKS Face Image Data Set: Collection Procedures and Progress Report, Technical Report MS-CIS-07-03, University of Pennsylvania, 2007.
• James Michael Alexander, MASKS: Maintaining Anonymity by Sequestering Key Statistics, Publicly accessible Penn Dissertations, Paper 32, http://repository.upenn.edu/edissertations/32/, 2009.

The MASKS Photo Set

A representative subset of these images were marked up using VisageMap, a facial image segmentation tool developed for the project. The fully annotated versions of selected MASKS photos will be made available for research purposes soon. The source code for VisageMap will also be made available under an open source license.

The details of how the annotation was done and what the first set of results to come from this research have been published as MASKS: Maintaining Anonymity by Sequestering Key Statistics.

Acknowledgements

• SF GRIDLOCK: YALE UNIV (Prime Y-02-0010) and NSF CCR-02-08972 (Originating Sponsor)
• University of Pennsylvania Research Foundation
• Olga and Alberico Pompa Professorship of Engineering and Applied Science
• NSF BIRT: ``CT-ISG: Biometric Identification Red Team (BIRT)'' NSF CNS-07-16552