A machine-vision algorithm that can detect and identify microexpressions--fleeting facial expressions that reveal deep emotions--with superior capability to humans has been developed and tested by researchers led by the University of Oulu's Xiaobai Li.
Li and his team first generated a database of videos displaying microexpressions in realistic conditions, by asking a group of people to watch footage designed to invoke strong emotions while hiding their feelings as they were filmed. The displayed emotions were then linked to the emotional content of the videos. The researchers next used a single frame showing a subject's face as a standard and compared all subsequent frames against it to ascertain how the expression changed, with any change beyond a certain threshold designated a microexpression.
"One major challenge for microexpression recognition is that the intensity levels of facial movements are too low to be distinguishable," the researchers note. They addressed this problem with an algorithm that "magnifies" expressions by identifying the parts of the face that move when an expression changes and distorting the face to move them further.
Lastly, the algorithm categorizes the displayed emotion as positive, negative, or surprise, a process it learns from the database.
Li's team says potential applications for the technology could include lie detection, law enforcement, and psychotherapy.
From Technology Review
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA