Symbolic and Subsymbolic Learning for Vision : Some Possibilities