If you are looking to recognize where a face is in an image, as opposed to matching faces across multiple images, there is actually a library that does this in conjunction with HTML canvass.
There is a demo on the developers site here. You can also download the source on github.
In my test's the performance was decent - but not blazing.
与恶龙缠斗过久,自身亦成为恶龙；凝视深渊过久,深渊将回以凝视…
