I am doing face detection application where every face detected is inside a rectangle. The rectangle always contains some background behind the face and now I need to extract as much of the face as possible.
Also the method should be fast so I've decided to extract just some basic shape from the rectangle, like an inscribed circle. This circle unfortunately also doesn't fit so well. The best shape I've tried right now is ellipse:
(2*width of the rectangle)/3 x height.
So right now I am looking for some better ratio between the width and height of the face maybe the best would be if you could point me to some research papers.