Presentation Review: Digital Human Face Recognition
Last night I went to the Austin Forum on Science, Technology & Society. The speaker was Dr. Shalini Gupta of AT&T Laboratories on Digital Human Face Recognition.I am actually fairly familiar with the subject, as I gave a presentation over all of the available biometric forms of identification at a BICSI Conference in 2008. I am not that deep into security systems as I’ve mainly been involved with telecommunications projects since graduating, but the convergence of security systems onto the IT network is becoming more evident year after year.Interesting points from Gupta’s talk/research:
- When you double the gallery/database size (number of people), the acceptance rate goes down 2 to 3%.
- When someone ages one year, the acceptance rate goes down 5%.
- The more highly variable a facial feature is among a population, the better identification metric it makes.
- Many facial expressions just bend the skin isometrically instead of stretching it.
The picture above links to research Dr. Gupta did while at the University of Texas at Austin. (This link will work too.)