It didn’t take long for us to get used to Facebook recognizing our friends in our pictures. Ever since, app developers enjoyed using this nifty little gizmo from Face.com to build facial recognition into their own projects. The Face.com API was free and could easily be integrated into applications to detect and identify human faces in pictures. However, Face.com was recently bought by Facebook, and has unfortunately, yet unsurprisingly, stopped providing this popular API. It’s been a few months since the acquisition, so let’s now assess what Face.com offered, and evaluate the remaining alternatives.
First, a quick look back at a study that measured the accuracy of the Face.com API. Three users provided six photos each and uploaded them to Flickr. Some of the photos were sharp and some were obscured; some contained groups of people, and some a person on their own. The photos were tagged with a random code, and the tag ‘London’, to see if the users could be recognized in a significantly larger pool of photos.
In the small, clearly defined photo set, facial recognition was fairly successful: in most of the photos the users were correctly identified. In the wider group, however, not a single correct photo was correctly identified. One of the photos returned with 12% confidence that this Roman statue was in fact one of the users:
So there was definitely room for improvement. Right from the start, Face.com’s API lacked the level of sophistication seen in the larger, but less accessible, Open Source Computer Vision Library (OpenCV). OpenCV is supported by a wide range of facial recognition algorithms, based on two separate models, and large online databases. It is, though, very much targeted at experienced web developers.
The only remaining option for casual app designers therefore is made by LambdaLabs. Their program provides basic face detection by identifying individual mouths and noses, and is thus currently a lot more limited than Face.com or OpenCV. There are, however, plans to improve facial recognition, in particular mood detection.
For those developers who are more than casual consumers of this type of software there also exists a wide range of high-quality proprietary facial recognition software. Certain packages include skin biometrics in the recognition parameters, but the majority aimed for use with security applications.
In general, facial recognition technology is still in the early stages of development, and is expected to improve considerably in the next few years. Cloud computing offers great potential in this area. Using cloud hosting for your application, means that users have access to virtually unlimited processing speeds, and the cloud provides the ample space that you would need, to store, for example, the immense data sets necessary for facial comparison. Here’s hoping more developers respond to the increasing demand for free, simple facial recognition APIs. Until then we’ll have to conclude that the Face.com API will be sorely missed.