November 19, 2014
Breakthroughs in image recognition technology may drastically improve image searches when machines can recognize people, objects, actions, and even the quality of photographs. Researchers at Google and Stanford University recently unveiled new software that can teach itself to identify the characters, actions, and settings of a scene in photos and videos. Photo sharing startup EyeEm has fine-tuned algorithms that rate photographs based on aesthetics.
These so-called “computer vision” technologies will likely lead to huge advances in image searches. Google, for example, currently has to rely on textual clues to categorize the content of the photo. With computer vision, image and video search results will be more accurate and surveillance cameras may be able to immediately identify a suspect or suspicious behavior.
“In the longer term, the new research may lead to technology that helps the blind and robots navigate natural environments,” The New York Times reports.
Fei-Fei Li, director of the Stanford Artificial Intelligence Laboratory, and his team approached the problem of image recognition by using software neural networks, which can “train” themselves to identify patterns in data. As the software processes more images and videos, it gets more accurate.
“I consider the pixel data in images and video to be the dark matter of the Internet,” Li said. “We are now starting to illuminate it.”
EyeEm has also developed its own computer vision software. The startup partnered with talented photographers to create a large collection of stock photography, and it needed image recognition technology to search and categorize the images. Their newest algorithms deal with the quality of the photo, so that the company can show users the best of their archives.
The scoring system looks at several aesthetic features, like the blurriness of certain objects or the placement of objects in each third of the photo, in addition to factors like human engagement, explains TechCrunch.