Consumer technology to pay attention to
Reading through my LinkedIn groups this morning, I saw a note from Martijn Kolenbrander, the vice president of business development for VMS/analytics firm VitaminD, that mentioned Face.com, a consumer-grade facial recognition solution. Martijn's message was from The IP Video Security Group, which is a group I'd highly recommend you join if you're involved in video surveillance in any way. I pointed my browser over to Face.com, where I learned that this application has been deployed in Facebook as "Photo Tagger" for automatically tagging photos of friends and family. Basically, this is consumer technology, designed for a consumer purpose that does what many high-end facial analytics software companies would hope to do. So much for the idea that facial recognition comes out of military labs first and spreads to the consumer channel. The same thing exists within Apple Computers' iPhoto picture management program. It has a facial recognition functionality that can learn who is who in your photos and automatically suggest them for identification (so that you can easily build a grouping of photos of specific loved ones).
The application of facial recognition in our professional security industry is clear. If you have a photo of a known suspect, let's call him "Suspect #1", you could have such a system look for all similar faces in your video and still images collection. Maybe Suspect #1 was at your bank one week ago cashing a check and casing the joint. Maybe he parked his Buick in your lot and you, by possibly recognizing him, would know to then find when he parked in your lot and write down that plate number as part of the case investigation. Sure, this is a theoretical example, but it's not that far off, and there are companies like 3VR and others who are doing similar, very important work right here in our industry.
While I've not used the Face.com application, it has some really great reviews. I've played with Apple's comparable solution in iPhoto. It's titled simply "Faces", and it works. While I have a relatively small collection of pictures in iPhoto, I've been reasonably impressed by the quality of this function. Sometimes it has been known to screw up horribly, suggesting a bearded man's photo might be another photo of a female family member (no women in my family have beards, mind you), but in general it performs reasonably well, and often better than "reasonably well".
Now before I laud such systems as ready to overthrow established facial recognition companies in our industry, I want to point out one caveat. When people take photos of their friends, they usually have their friends looking directly at the camera; they have the flash turned on to get a nice picture, and they're using cameras with many megapixels. In our industry, video surveillance cameras rarely have the resolution of your standard consumer-grade point-and-shoot digital camera. Neither do we often control the lighting through an "auto flash fill" feature on these consumer still image cameras. And criminals are usually trying to keep their mugs out of the view of cameras. So, the industry technology has to work at least three times as hard as consumer technology.
We're getting better at giving them higher quality video, though. Camera placements are going to more head-level locations (cameras at eye level at bank teller windows, at eye-level at retail door exits, etc.). Every year, we progress toward higher resolution cameras that have better wide dynamic range for dealing with bad lighting.