Program Developed to Learn Everything It Can
We all know how frustrating it can be to search for something on the Internet and simply be unable to find it, or be presented with so much information you do not know if you are looking at all of the relevant information. If you are lucky, you can figure out how to exploit search engines to get what you want, but it would still be nice if we could get some help. Researchers at the University of Washington may soon be providing that help with their Learning Everything about Anything (LEVAN) program.
This program is the first fully automated program that teaches itself everything about any visual concept. By providing LEVAN a query, it will start searching through millions of books on Google Books to find every occurrence of the concept in the text. An algorithm then identifies and removes words that are not visual, to arrive at a list of relevant phases. Those phrases are then used to perform an image search, which LEVAN then analyzes for similarities to recognize the images associated with the concept. After all of that, it is able to provide a user with a comprehensive list of images that can be easily browsed.
As you may be able to guess, the process of LEVAN learning a concept is not quick and can take up to 12 hours. The researchers are working to increase processing speed though, and they also intend to make it open source, for educational tools and for others in the computer vision community to use.
Source: University of Washington