On the introductory page, I claimed that the ganglion cell layer is a bottleneck. Let us look at this analogy, see if it is appropriate and, if not, come up with a better one. A bottleneck is a common analogy to indicate that there is a point where less "stuff" can pass. In a real bottle, the fluid is held up exiting through the neck of the bottle because it is thinner than the rest of the bottle. However, a bottleneck literally eventually lets everything through, more slowly perhaps but everything get through in the end. You might even have some turbulence if there is too much liquid trying to get through.
Somehow that is not what I am trying to get across. I think some of the visual pattern of light does not get out of the retina. Yes, the ganglion cell layer is a bottleneck in that it is thinner, but not everything is allowed through. Perhaps a sieve is a better analogy. A sieve, in a general sense, only lets some "stuff" through and not other stuff. In a real sieve it is that only that part of the flour, or whatever is being sifted, that is small enough to get through will be allowed through. In applying this analogy, it is not necessary to limit the sifting to size. We might conceive of a sieve that sifts basing on shape or other characteristics of the material. So a sieve is different from a bottleneck. In a sieve, something can get left behind which seems to be more to the point. It is not a slowing down of information as much as a selection of what gets transmitted. Moreover, the selection is not random us meaningful and done in a way to get as much through as needed. If we apply selective pressures to various sieves, then those that let useful information through will give a survival advantage.
In addition, to the proper analogy for the functioning of the ganglion cell layer, here are some of the background ideas that I am considering in developing this model. In time I hope to flesh out these ideas and describe and explain them in detail.
The mind is a pattern matcher (Hofstadter, 1980; Shepherd and Cooper, 1982)
One of the main purposes of the retina is information reduction. There are ~7 million cones and ~120 million rods per eye and only ~1.1 million ganglion cells. Compression might be a better word.
There are at least two types of receptive fields. I am going to use the Enroth-Cugell and Robson (1966, 1984) distinction of X and Y for reasons I hope to explain. These two receptive fields have different response properties which are very important to how we reduce our perception.
The visual system is able to support our seeing because in some sense it behaves isomorphically with the real world. I have a discussion of this issue here.
There is emerging evidence that these receptive fields are dynamic (Ramachandran, 1993). That is, the region that they respond to on the retina is not constant.
There are some important truths to the gestalt and Gibsonian point that our senses work to extract invariance, that is in this sense relevant information (Gibson, 1979).
Change in the stimulus is very important for our visual front end. Here are some examples of the stimuli that show this specialness of change: Mach Bands, Craik-Cornsweet, Minimal Contours, Pseudo-stabilized images.
The eye is never stable. Even during fixation the eye has tremor that moves it across the image. For most models this is a problem so they leave it out. I believe that the visual system actually takes advantage of this fact.