From: The World In Your Head by Steven Lehar

Fig. 5.6 The noisy retinal image is used to construct a three-dimensional model of perceived surfaces and illuminants by reification. That perceptual model in turn is used to create a two-dimensional projection of the modeled scene. The projection is then compared with the retinal image, and the differences between them are used to update the modeled scene, in order to minimize the differences between its predicted projection and the retinal input.