Viola Jones Algorithm: Scaling of Detector

edit

I am actually implementing the whole algorithm from scratch. I am able to detect the faces of particular size from image by resizing it to appropriate value. But how exactly the scaling of detector has to be done to scan the faces of different sizes? Position and size of feature can be multiplied by scaling factor, there is no problem in that. But what about the threshold, which makes the criteria for a particular (weak) classifier to qualify a window as face or non-face.