We do not see objects as such; we see shape, surfaces, contours, and boundaries, presenting themselves in different illumination or contexts, changing perspective with their movement or ours. From this complex shifting visual chaos, we have to extract invariants that allow us to infer or hypothesize objecthood. It would be uneconomical to suppose that there are individual representations, or engrams, for each of the billions of objects around us. The power of combination must be called on; one needs a finite set or vocabulary of shapes that can be combined in an infinite number of ways, much as the twenty six letters of the alphabet can be assembled (within certain rules and constraints) into as many words or sentences as a language ever needs.
Mark Changizi and his colleagues at Caltech examined more than a hundred ancient and modern writing systems, including alphabetic systems and Chinese ideograms, from a computation point of view. They have shown that all of them, while geometrically very different, share certain basic typologies...Changizi at al. have found that similar typological invariants in a range of natural settings, and this has led them to hypothesize that the shapes of letters "have been selected to resembled the conglomerations of contours found in natural scenes, thereby tapping into our already-existing object recognition systems.
(Image from Magic Forest by Andrew Carnie)