Previous Image « » Next Image Click to Enlarge Click to Enlarge

Basic Training About Face!

The story of face recognition technology begins back in 2005, when Nikon released the Coolpix 5900, a mid-range point-and-shoot still camera with what it called a Face Priority mode. The camera would use built-in algorithms to examine a scene and identify human faces. It would then adjust contrast, color balance and exposure to "properly" expose the faces. Other still camera manufacturers followed suit, and today even high-end DSLRs (digital single-lens reflex cameras) ship with the feature. Something this exciting wasn't about to remain solely in the still photography world. By 2008 Sony, Panasonic and Sanyo had all incorporated face-detection features into their current batch of video cameras. But what is face detection? A gimmick or a useful tool? Let's take a look.

Our own eyes are automatically drawn to other human faces. When we're walking down the street or glancing at images, our eyes leap to faces and our brains instantly attempt to determine not just whether or not we know this person, but also mood, attractiveness or whether this person may pose a threat. It all happens in fractions of a second, and most times we don't even know that we're doing it. It makes sense then that, when we look at video, we'll also seek out faces, so, in recording video, we want to make those faces look their best.

Often in life, people and lighting are not as a director would place them - huge windows may backlight a bride and groom, bright sunlight may make dark eye sockets or overhead fluorescents may throw a green hue onto someone's skin. Ever-changing lighting and distance conditions have made auto-focus and auto-exposure indispensable in modern camcorders. Face detection is no different. It will identify a face, track the face if it moves across a frame and then perform a series of tasks, such as selecting a focus point or brightness and contrast.

Most face detection algorithms recognize not just one, but multiple faces. Usually they consider the face closest to the center the most important or primary face; they notice others behind it or on the periphery, but focus and image adjustments will follow the primary face. If the primary face leaves the frame, the software will choose a new primary face. Many of the current crop of cameras will pick five or more faces (if they're available).

Though sometimes used interchangeably, the terms facial recognition and facial detection are actually two different technologies. While your camcorder can pick a face out of a crowd (facial detection), it doesn't (yet) have the ability to search a database and identify the people in the frame (facial recognition); it knows only that there's a face in the frame - it's completely clueless as to whose face it is. Companies like Betaface (among others) have sophisticated methods of identifying specific people, based on techniques similar to this, but that take more accurate measurements of the distance between eyes and mouth, the width of the mouth when open and other things, such as hairstyle and whether the person is wearing glasses. It can then compare these measurements to a database and look for matches. The idea that your camcorder will someday be able to do something like this on the fly is not farfetched. Already social networking sites such as Facebook allow viewers to manually tag photos with the names of people in them and then link them to a pool of other photos. This suggests that the desire for automatic recognition (and tagging, by extension) is there, and now it's merely a matter of the technology catching up.