Bellevue-based Eyealike recently introduced VisualAd, an image and video-based contextual ad platform--Eyealike VisualAd recognizes facial features, skin colour, gender, and age, and discerns logos or product images. This gives them a means to offer highly-targeted ads by indexing, filtering and classifying images and videos rather than using keywords for the targeting purposes. Admittedly, this concept makes me feel slightly uneasy, but posting pictures to social networking sites is done so at the expense of the user's privacy anyway.
Using this type of “dual serving” technology offers these sites twice the impact on ad relevancy. Eyealike President Greg Heuss calls the technology "dual serving"--he insists that social networking sites are interested in what types of images you post, what you look at, and what they tell the SNS about you. I was curious about how VisualAd will infer information about a user based on photos that will undoubtedly feature other people. CEO John Hafen argues that Eyealike first "tells" the technology what given photos are. Then the software "gets smarter with time and with more and more images to view. After a while, it becomes self aware"--learning and then self-teaching, presumably, as it learns to cluster images based on content.
Eyealike may have an early advantage through the accuracy of their recognition algorithms, which are apparently at 90% according to Hafen. VisualAd uses data from all of a user's profile on their user-generated content site. It extracts a certain probability that a target object is in each individual photo, and then if that target object (or objects) are in multiple photos the probability increases that the user has that object (or is at least in close proximity). This will work the same whether the object is a baby, a Golden Retriever, a Subaru, or Mount Rushmore.
Hafen attests that "What will put us ahead of the pack is our accuracy, how we apply it, and the basic scalability of the system. We built it from scratch with super-scalability in mind." Scalability will be important, because Eyealike is looking to market directly to social networking sites. Sites like Facebook and MySpace will likely be eager to monetize user-generated photos.