A new angle on face recognition

TechCrunch points to a new web-based face-recognition service called Polar Rose. With similar goals to those of Riya, Polar Rose wants to help you recognize or locate faces in online images. Riya didn’t have much success at this and has changed their business model and my guess is that Polar Rose won’t be far behind.

Why?

This is a hard problem. Recognizing faces in a handful of images with a couple hundred faces is one thing–doing so with the myriad of online images is another. Polar Rose is going to take advantage of people tagging images in order to help train their engine, which makes sense, but this still doesn’t change the fundamental problems of the quality of the training set. It’ll be interesting to see how it all works out or if they can find a sweet spot in how to use the technology. I’m skeptical though. They may be able to leverage a patent or two later along the way, however.

To me, if people want to change the world when it comes to face recognition they’d focus on image streams, such as those from webcams, (possibly) cellphones, conference systems, and other digital cameras. A webcam on your laptop, that watches you day in and day out has a better chance of building up quality recognition vectors when shared across systems–whether they are the computers in your house, office, belong to a group of friends, or people scattered across the Internet.

Unfortunately, few of these devices as well as the firmware or software that comes with them are set up for developers to push this type of technology. For instance, given a webcam, what’s the best way to share the image stream so that additional analysis can be performed without impacting any other apps? Over the years, there have been ways to do this, but most are a kludge. The technology scape hasn’t been focused on helping recognition grow. Trying to leverage discontinuous images on the Internet seems like a tough way to go–outside of leveraging humans tagging most of the images for you.

All this being said, there are a few cases where I can see people wanting to look up others or themselves online using a face recognition service.

For instance, bloggers might want to subscribe to an RSS feed that gives a link to every new (or existing) picture of themselves on Flickr.

Or someone might ask: Did my ex-boyfriend/girlfriend post a video or pictures of me online? A potential list of hits might be of interest to someone to go through.

Or you want to build up a photo-album for a friend or relative and you go online to collect up images you might not otherwise have.

I’m sure there are others. But to me, there’s still much ground work to be done at the device level, whether its through additional sensory data, additional processing, or additional human interaction.