Describir: Invariant models of vision between phenomenology, image statistics and neurosciences