A Canadian computer science grad is mapping the way the human brain works to technology that will power a search engine for visual images to be launched next summer.
The University of Ottawa said Master’s student Kris Woodbeck was working with the government’s Technology Transfer and Business Enterprise (TTBE) office to secure a patent on his approach, which will form the intellectual property for a startup devoted to image search. Woodbeck won the school’s Innovator of the Year award last week.
Woodbeck said he has already created a prototype of the search engine based on his patent, which apes the way the brain processes visual information and tries to take advantage of currently-available graphics processing capabilities in PCs.
“The brain is very parallel. There’s lots of things going on at once,” he said. “Graphics processors are also very parallel, so it’s a case of almost mapping the brain onto graphics processors, getting them to process visual information more effectively.”
Although the full details of his company’s business plans have not been worked out, Woodbeck said he sees the potential for his patent far beyond just generic object recognition. Besides the publicly-available search engine, there could be opportunities to licence the intellectual property for medical and military markets. Biometrics sectors such as facial recognition may be another possibility, he said.
“All the major (search) players don’t really look at visual content,” he argued. “Their results aren’t very good. You’ll see a search engine, for instance, that uses the colour of the picture to classify it. It’s very new field. It’s in its infancy because everyone is using metadata. Nobody’s looking at the content, because it’s so slow to process images.”
Guy Creese, an analyst who specializes in search with Midvale, Utah-based Burton Group, said vendors are struggling to find the right kind of artificial intelligence to extract what’s in an image and create the right kind of metadata.
“In text, you’ve got a lot of metadata compared to images – company names, location, et cetera,” he said. “For images, it might be when you took it, with what camera, with what exposure, that’s about it. Then you’re stuck with a red barn in rolling hills and I might know it was taken in California, but no one else does. How do you surface that metadata so it becomes much more searchable?”
The real problem is that indexing such content becomes manually intensive, Creese said. For a company that sells photos, it might be worth having human beings doing that kind of work, but for most organizations there needs to be a way to automate the processes, he said. Image search will also increasingly dovetail with video search, he added, which may offer more potential for tagging.
Woodbeck said he has been testing his search engine technology on academic data sets that include between 60,000 and 100,000 images, trying to classify them according to certain labels. He said he will be using Web crawlers to start indexing images so that his search engine will have a critical mass of content prepared at the time of launch. The TTBE will be assisting with some of the business development aspects of the startup, he added.