Natural Sciences and Engineering Research Council of Canada
Symbol of the Government of Canada

Common menu bar links

Past Winner
2008 Innovation Challenge Award

Kris Woodbeck

Object Recognition with Neuroscience and Graphics Processors

University of Ottawa

The odds of finding just the right image on the Internet could soon become a lot better thanks to a new type of image search engine developed by Kris Woodbeck at the University of Ottawa. His invention has earned him a third-place prize in NSERC's 2008 Innovation Challenge.

Current image search engines rely on text “metadata” tags that are attached to image files in order to return results based on entering key words. Until now, attempts to create systems that could search based on the visual characteristics of the image itself have been cumbersome and inaccurate.

Woodbeck set out to simply improve the speed and accuracy of existing models for object recognition, but made a key discovery along the way that opened the door to a completely new approach. Instead of designing his system using conventional computer processors, he turned to the latest generation of computer graphics processors, and found that these have one key similarity to the human brain: the ability to process fragments of visual data simultaneously. Taking advantage of this characteristic enabled him to create a system that comes closer to mimicking a human being's finely tuned ability to identify objects.

The new “Biologically Motivated Search Engine” not only does a better job, but works quickly enough to enable a visual recognition system to compete with text-based image search engines. Woodbeck estimates that his software running on graphics processors can do the visual processing of 100,000 conventional processors.

In addition to improving Internet search engines, Woodbeck's system could help retailers by making it easier for consumers to find their wares. The high popularity and rapid growth of image searches on the Internet means that there is a large potential market for a new, more effective approach. After successful tests of the system at various index sizes, the system has now been scaled up to a search engine that will catalogue hundreds of millions of images.