Billions and billions of images exist online. Despite some search engines' best efforts, organizing all these pictures in a useful way is an arduous task.
A UW-Whitewater professor is working to teach computers to recognize images and group them based on content -- skills that human beings take for granted.
Lopa Mukherjee, assistant professor of mathematical and computer sciences, received a $231,313 grant from the National Science Foundation to advance computer vision technology.
Mukherjee noted that, based on millennia of evolution, human brains can easily recognize familiar forms such as faces and landmarks and relate them to images they have seen in the past.
"Computers are not that lucky," Mukherjee said.
Mukherjee and her students will work to develop software that will allow a computer to recognize subjects, such as the Eiffel Tower or the president of the United States, and, based on similar images it has dealt with before, link them to other pictures of the same subject matter or structure. At the same time, the computer will be able to sense commonalities within a collection or album of photos.
For example, when a computer comes across a photo of the Eiffel Tower, it could identify that particular shape, recall other photos with that same shape tagged as "Eiffel Tower," and be able to name that image while categorizing it with similar photos.
This would mean more than improved Internet image searches, Mukherjee said. People could create 3-D models of landmarks and locations without assembling the photo reference material themselves. This would be a major advancement for video game developers, special effects technicians, and even students enrolled in online classrooms in Second Life.
Mukherjee's project is called "analyzing subspace structure for group level image understanding." Her three-year grant was awarded by the Information and Intelligent Systems Research in Undergraduate Institutions Program of the National Science Foundation.