Cavanagh comes up with some interesting models of how our top-down vision processing might work. However, I find it important to point out that his model of prototype matching is merely a theory. There is no real hard evidence (at least not that I know of) that we internally have a designated prototype/2D template matching system. Surely, our process does externally appear to be some sort of matching model, but internally it might contain much more subtle underlying neural processes than just the algorithmic symplification of visual prototype-matching.
And even if we happen to use a prototype model, the question is how much of the system is purely vision-based and how much of it is connected to rather non-visual semantic knowledge (and whether these are really separate entities). We might for instance be able to visualize and identify things that we have never seen, but of which we have heard a purely linguistic description. I believe that our top-down processing is a lot more complex and interconnected with more abstract/semantic world knowledge than Cavanagh's purely visual 2D template model describes.