Setting the Computer's Sights
One challenge computer vision has to address is being able to match the capabilities of the human eye. Dr. Tal Hassner of the Department of Computer Sciences at the Open University and his team have created a program which enables computer recognition of faces with limited indicators, successfully competing with the leaders in the world.
Dr. Tal Hassner, along with his colleagues from Tel Aviv University, Dr. Lior Wolf and his student Yaniv Taigman (also from the Israeli start-up face.com) decided to tackle one of the most difficult problems computer vision has to address: Are these pictures, with varied backgrounds and from different perspectives, representative of the same person?
Using the internationally recognized benchmarking figures of the University of Massachusetts, Tal and his colleagues have scored 86% reliability, 12 points better than the original, state-of-the-art program from 2008.
To understand why this question poses so many difficulties, and Tal's team's results are so impressive requires a deeper understanding of how visuals in general and faces in particular are interpreted by the computer and by humans.
|
Academic Commandos
E-reader
Walking in Their Footsteps in Ashdod
Dance with Your Fears