Communications of the ACM
How to Pick Out a Face in the Crowd
Five facial feature points remain unchanged by different expressions, researcher Dacheng Tao says.
Credit: Thinkstock
A computer scientist from Sydney's University of Technology and colleagues have addressed the accuracy limitations of existing facial recognition technology. Dacheng Tao, from the university's Center for Quantum Computation & Intelligent Systems, describes the team's new algorithm as a major breakthrough.
"Since computers calculate measurements on the face numerically, it's widely acknowledged that different facial expressions — if you smile or are angry, if the image is not frontal, even if you wear make-up and glasses and if the lighting is different — can affect these statistical features when comparing photos with one another," Tao says. The new algorithm is designed to measure the center of the eyes, the peak of the nose, and the corners of the mouth. Tao notes the five facial points are strong, stable, and change little in different environments.
Tao says the algorithm is based on multi-modal deep learning. The process involves taking the points from a two-dimensional image and extracting robust statistical features for subsequent recognition of a subject.
From Sydney Morning Herald
View Full Article – May Require Free Registration
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found