Friday
May072010
New Technologies Improve Object Recognition Systems
Friday, May 7, 2010 at 12:55PM
Researchers at MIT and UCLA have developed new techniques that should make object recognition systems easier to build and enable them use computer memory more efficiently.
A conventional object recognition system, when trying to discern an object, will generally begin by looking for the object’s salient features. A system built to recognize faces, for instance, might look for things resembling eyes, noses and mouths and then determine whether they have the right spatial relationships with each other.
The design of such systems, however, usually requires human intervention: a programmer decides which parts of the objects are the right ones to key in on. It also means that a system designed to recognize millions of different objects would become unmanageably large.
The researchers' system learns to recognize new objects by being “trained” with digital images of labeled objects. But it doesn’t need to know in advance which of the objects’ features it should look for. For each labeled object, it first identifies the smallest features it can — often just short line segments. Then it looks for instances in which these low-level features are connected to each other, forming slightly more sophisticated shapes. Then it looks for instances in which these more sophisticated shapes are connected to each other, until it’s assembled a hierarchical catalog of increasingly complex parts whose top layer is a model of the whole object.
The system saves memory because different objects can share parts. That is, at several different layers, the parts catalogs for a horse and a deer could end up having shapes in common. Wherever a shape is shared between two or more catalogs, the system only stores it once.
A conventional object recognition system, when trying to discern an object, will generally begin by looking for the object’s salient features. A system built to recognize faces, for instance, might look for things resembling eyes, noses and mouths and then determine whether they have the right spatial relationships with each other.
The design of such systems, however, usually requires human intervention: a programmer decides which parts of the objects are the right ones to key in on. It also means that a system designed to recognize millions of different objects would become unmanageably large.
The researchers' system learns to recognize new objects by being “trained” with digital images of labeled objects. But it doesn’t need to know in advance which of the objects’ features it should look for. For each labeled object, it first identifies the smallest features it can — often just short line segments. Then it looks for instances in which these low-level features are connected to each other, forming slightly more sophisticated shapes. Then it looks for instances in which these more sophisticated shapes are connected to each other, until it’s assembled a hierarchical catalog of increasingly complex parts whose top layer is a model of the whole object.
The system saves memory because different objects can share parts. That is, at several different layers, the parts catalogs for a horse and a deer could end up having shapes in common. Wherever a shape is shared between two or more catalogs, the system only stores it once.
Reader Comments