Images account for nearly 50 percent of the average Facebook News Feed’s content. Assuming Facebook is a proxy for the entire Internet, then roughly half the pixels on the Internet are black boxes to ad targeters.
While keywords have created multibillion dollar ad businesses like Google’s, the search giant’s famed crawlers can’t detect what objects appear in an article image or user-posted photo, preventing the possibility of serving a Coke ad to someone who just looked at picture with a soda can in the background. In fact, no one has been able to crack images—though fissures are starting to show.
While many image-recognition experts agree that true computer vision—the ability to index images as easily as text—is a decade if not a generation away, a number of companies are setting the foundation for that future. WPP’s ad tech outfit 24/7 Media has partnered with visual advertising startup TripleLift to turn clicks on brand photos shared on social platforms like Facebook, Pinterest and Tumblr into ad retargeting signals.
“If you share a photo of a car on Facebook and I click on that post of a new Aston Martin, I’m now added to the Aston Martin segment where I can be targeted by Aston Martin or other [auto brands],” explained Ari Lewine, TripleLift’s co-founder. Rob Schneider, 24/7 Media svp of corporate strategy and platform development, called the retargeting product 24/7 SociAble, “the Criteo killer.”
However, the fact that the retargeting product relies on clicks on a brand’s own images underscores computer vision’s obstacles (Lewine made clear that computer vision isn’t core to TripleLift’s technology). Because a computer would need to process petabytes of images to be able to instantaneously understand an image’s contents, image-recognition company Luminate augments its computer vision software by processing the text on a page, partnering with companies like Getty Images to connect a photo with information in Getty’s database and even employing humans to fill out more precise data.
“Identifying a Starbucks logo can be done with software because it’s a predictable, recognizable set of pixels,” said Luminate chief revenue officer Chas Edwards. “If you’re L’Oréal or Aussie and want to market a shampoo product when someone’s looking at a photo of a woman with highlighted hair, that requires more finesse.” Finesse like the kind Google has used to understand textual information.
Google Image Search’s visual stockpile and the company’s image-recognition crowdsourcing product Google Goggles position it well to realize true computer vision and monetizing the data it unlocks. But Facebook made two acquisitions last year that could move it ahead of Google. First it bought Instagram, and with it the 40 million images the photo-sharing platform’s users upload every day. Then it snagged Israeli startup Face.com, whose facial recognition technology could be applied to broader object recognition, such as those in the more than 219 billion photos on Facebook.
“Facebook’s acquisition of Face.com ratifies our position that visual content is super important to what these guys do next in monetizing the assets they have,” said Jamie Thompson, CEO and co-founder of image recognition startup Pongr.