Communications of the ACM
Seeing the Forest For the Trees
Credit: Christine Daniloff / MIT News
New object recognition systems that deconstruct images into ever smaller elements, using methods developed by researchers at the Massachusetts Institute of Technology and the University of California, Los Angeles, should be much more efficient than previous systems and may yield insights on brain behavior.
The researchers have developed a system that learns to recognize new objects by being "trained" with digital images of labeled objects. For each labeled item, the system first identifies the smallest elements, and then seeks instances in which these elements are interconnected into slightly more complex configurations. The system continues to search for instances in which shapes of ever increasing sophistication are linked together until it has put together a hierarchical catalog of increasingly complex components whose top layer is a model of the entire object. The system then sifts through its catalog from the top down, weeding out all redundancies. Memory is saved because different objects can have shapes in common, requiring only once instance of memory storage.
From MIT News
View Full Article
Abstracts Copyright © 2010 Information Inc., Bethesda, Maryland, USA 
No entries found