I would recommend Vision by David Marr (1982). It studies early vision relating neurophysiology to algorithms and assumptions about the physical world. The key point it's making is there are three different levels of explanation when studying how the brain achieve s a certain task, and one should try to keep them separate if one wants to achieve clear understanding. The sort of thing he is criticising is fuzzy thinking like "deep learning is inspired by how the brain works". So for stereopsis he goes from psychological experiments looking at visual illusions and random dot stereograms ( magic eye?) which illustrate the assumptions about the 3d world that allow us to recover depth from 2 2d images, to possible neurophysiological implementation s.