Computer-vision models designed to reproduce the capacities of the human brain may need to be adjusted, suggests researchers from the Weizmann Institute of Science and the Massachusetts Institute of Technology.
The team enlisted thousands of participants from Amazon's Mechanical Turk to identify a series of images, and compared their scores to computer models. Humans were much better at identifying partial- or low-resolution images, but nearly everyone also stumbled at the same exact point of fairly high loss of detail.
The researchers say computer models identify images by moving from simple features to complex ones. However, the human brain does this while simultaneously comparing the elements in an image to a sort of model stored in its memory banks.
The research suggests there is a minimum amount of information an image must contain for recognition to occur, and a person's brains may contain something that is tuned to work with a minimal amount of information. Incorporating this elemental quantity into current computer-vision models could improve their sensitivity, according to the researchers.
From Weizmann Institute of Science
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found