A cognitive psychology experiment has revealed key differences in how humans and computers see images, says Weizmann Institute of Science professor Shimon Ullman.
He says the results could help improve computer-vision algorithms and lead to the development of artificial intelligence that learns to understand the world the way a growing toddler does.
The experiment found human vision outperformed computer vision, but human recognition dropped suddenly when slight changes make images too small or fuzzy to identify, while computer algorithms did not show a similar "recognition gap."
Similar to human vision, computer-vision models rely on a bottom-up approach that filters images based on the simplest features possible before moving on to identify them by more complex features. However, Ullman says the human brain also works top-down, comparing a standard model of certain objects with a particular object it is trying to identify. He believes the top-down approach could be used to improve computer models and algorithms.
Ullman and colleagues have received funding to pursue this theory via a "Digital Baby" project grant provided by the European Research Council.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA