Neural Network Structure for Face Detection

See more:

artificial-intelligence

So i've been reading a lot about neural network for face detection, and all of the paper i've read suggested a neural network with an input layer of 20x20 image and 3 hidden types of hidden units:
4 which look at 10x10 pixel subregions,
16 which look at 5x5 pixel subregions,
and 6 which look at overlapping 20x5 pixel horizontal stripes of pixels.
something like this http://i61.tinypic.com/2w3x6ko.png.

I want to know why this structure is used, can anyone explain this in detail?
I'll be very grateful for any answer.