Nedko

Research has shown that 70% of the pictures we take feature people. Face detection in software in digital processors can greatly enhance the quality of images and help describe the content. Why not? While humans can easily identify objects in a photograph it takes a lot more to teach a machine how to do that and most of all …do it well.
First off it’s very important not to confuse the term “face detection” with “face recognition”. Although they sound very similar they are two completely different things. Face recognition is the ability to identify or verify a person from an image or video by facial features. Face detection is the ability to determine the location and size of human faces in images and video. Needless to say face detection is an important prerequisite for face recognition and as such plays an important role in recent research as government agencies are struggling to provide safe, inexpensive, and reliable security.
Before I jump into more details about the process and the techniques used I wanted to share my pure amazement when I found out that my digital camera’s face detection is racist. That’s right – while taking pictures at a party I noticed that my Canon camera cannot detect faces very well when the subject’s skin color happens to be other than white. Furthermore I realized that this is true even for mixed pictures containing different people. Acknowledging this interesting fact I was not very surprised when I started to look in the techniques used for face detection.
Digital cameras use face detection technology to adjust the focus and exposure of subjects and also allow the photographer to zoom in the faces while shooting. This is accomplished by dynamic scanning of the image and detection of patterns by the digital processor. There are many different techniques used in detection – some easy and some hard. Of course the easiest way is when we have images in an environment with plain monocolor or controlled background. Another approach is trying to recognize the certain facial features: head generally round shape, nose, eyes, chin, mouth, and ears. A similar approach detects objects in motion – the face is almost always a moving element in photos or videos. And if you thought that color needs to be somehow involved in our mystery you guessed right. Another very wide spread face detection technique is finding faces by color. This is generally a two step process which first finds groupings or regions likely to contain skin and then tries to analyze them and extract information to identify a face. Here come the disadvantages: this method requires color images, does not work very well with various kinds of skin color and the performance varies under changing lighting conditions.
The conclusion: don’t be surprised next time your camera favors some of your friends over others – you know why.

Filed under:
Development, General by Nedko