iSGTW Feature - BXGrid

Feature - BXGrid: ushering in the future of security

A page in the BXGrid portal that allows the user to check the correctnes of new ingested face images against older data. (Click on image for more complete view)
Image courtesy of BXGrid.

Facial recognition systems and iris scanners may seem straight out of a science fiction movie, but such biometric technologies could be the future of security - from ATM identity verification to improved airport safety. Currently, biometric identification technology requires high-quality images obtained under near-perfect conditions. The challenge is to design systems that can accurately identify people from imperfect images.

To improve identification techniques, biometrics researchers must run computation-intensive tests involving terabytes of image and video data. The Cooperative Computing Laboratory (CCL) and the Computer Vision Research Laboratory (CVRL) at the University of Notre Dame constructed the Biometrics Research Grid (BXGrid) in order to provide enough computing resources.

BXGrid assists with the entire biometrics research process - from collecting data to generating publication-ready results. The system consists of three components: a database, an active storage cluster of 16 servers, and a computing grid of 500 CPUs from the Notre Dame portion of Open Science Grid. Users can browse and manage data through a convenient Web portal, and the system's command line tool allows them to run customized tests.

"BXGrid is replacing and automating what has historically been a human-centric and error-prone data management process so that assembly of data sets, error detection and correction can proceed much more quickly, generating a higher-quality database as output," said Patrick Flynn, director of the CVRL.

Page in the BXGrid portal that allows user to browse the selection of iris images. (Click on image for more complete view)

Image courtesy of BXGrid.

CVRL researchers collect thousands of images and video clips of volunteers on campus and use them to test new facial image or iris scan identification algorithms. The BXGrid database indexes all the images and manages the accompanying metadata that describes the subject, the camera, and the conditions for each. Researchers can query the database and easily identify images that match their desired criteria.

To allow for computerized comparison, the images must be converted to a computer-readable format. For example, facial images become structured meshes, and iris images become iris codes, very compact descriptions of the fine structures in an iris. For video, researchers must first select the "best" frame, and then convert the image. Each image is then compared to every other image according to the researcher's test algorithm that determines which faces are similar.

Transforming a 1 MB image into a mesh takes about 1 second, as does comparing two meshes against each other. Comparing every pair in a set of 4000 meshes would take 185 days on a single CPU, but with BXGrid, it can be done in a day. Comparison of iris codes is about 50 times faster than structured meshes.

"In the past, processing just 100 images was considered a big result," said Douglas Thain, project director at the CCL. "In its largest run so far, BXGrid has enabled the processing of 60,000 images at once - the largest biometric experiment ever run on public data, by several orders of magnitude."

-Amelia Williamson, for iSGTW