Arizona-based img surf, LLC is an angel-funded startup that was formed in July 2007 to develop Mugr, a facial recognition search site designed to show off their developer platform that let's outside sites integrate image based search into their services. The site opened last month in limited beta. We were able to get our hands on some invites for Read/WriteWeb users, so the first 100 people who visit this link will be able to join Mugr and try it out for themselves.
Like the Eyealike, which we profiled on Monday, Mugr is more of a tech demo for the platform than a really useful destination site. On Mugr, users are encouraged to enter at least 3 clear, head on face shots to train the engine to understand who they are. They then fill out a painfully simple user profile that consists of links to various social networking or other web sites and tags that describe them, and can form connections with other users.
Once the site scans your photo and extracts various metrics from the photograph to learn your face, users can search for you by uploading other photographs of you. Mugr's technology uses a variety of methods to extract facial information and determine the boundaries of your face, which increases the accuracy of the algorithm. The image below illustrates the kind of things Mugr does to gather data on what your face looks like (on the left, the Mugr technology has shaded in red the area it believes is a face, on the right, it has created a wire-frame of facial features/boundaries).
The social image search site Mugr has built at mugr.com to show off their technology doesn't offer much in the way of useful functionality. You can input your likeness into the site and then your friends can search for you using photos of you that Mugr hasn't seen, including pictures sent from mobile phones. Due to privacy issues, Mugr is restricting search results only to users who are marked as connections, which severely limits the usefulness of the site. But the point isn't to be a destination, it is to show off their technology and promote their API.
Indeed, Preston Lee from Mugr told me that they are mainly interested in getting their API in the hands of developers right now as they continue to develop that end of their service. Mugr's API is currently at version 0.0.1, and will be free (for reasonable use) to developers. The company has also set up an API discussion group at Google Groups, and released Rmugr, which is a Ruby implementation of the API.
Mugr's platform intends to provide the facial image recognition, tagging services, and search layer for companies wishing to do visual search. What the technology is used for by developers is fairly limitless, and I got the impression that Mugr hopes to work with developers to build out their end to suit the sort of things developers want to make. I was shown an example tie in with the Flickr API that tagged Flickr photos in which faces Mugr recognized appeared and then linked back to that user's Mugr profile. Not the most practical application, but it hinted at the potential for future automatic tagging apps that could save people a lot of time when sorting photos, for example.
Like other image recognition technologies, Mugr is sure to have the same sort of allure to security and military intelligence firms. But it will be really interesting to see where developers will take this technology. What sort of apps would you like to see created on top of the Mugr API? And remember that there are 100 beta account invites waiting for R/WW readers at this link. Once signed up, users will receive additional invites to pass out.