I've been thinking of creating a pixel-array containing a binary image of the polygon shape and applying it in a conjunction operation with the source pixel array. 

Maybe there's a more efficient/smarter way of doing this  ?