Towards an understanding of natural constraints on visual representations