Title: Factorization of form and motion from time-varying natural images.
Keywords: natural images, 3D form, motion.
Abstract: During natural vision, images result from the 3D environment being projected onto the retinae as the eye, head and body move about in the world. Despite the fact that these images are highly unstable, our percept of the world is stable and assimilates information across multiple views. What are the computational problems that must be solved in order to do this? What are their neural substrates? Here I shall argue that the computational problem takes the form of a factorization, in which 3D form and motion are estimated jointly rather than independently in separate ‘what’ and ‘where’ processing streams. I shall present computational models for how this may occur in visual cortex, and testable predictions that arise from these models.