Interesting Article about the Eye: Biology, Psychology, Art

[Ishtvan]

http://physicsworld.com/cws/article/indepth/45840

 

This reminded me about our discussions of whether to model the AI's retina vs. the rest of their eye, and then introduce eyeball movement patterns - it's probably a good thing we didn't try to implement that with a simple model, because people are still discovering more about it nowadays. But this article argues about why we shouldn't use uniform, 2-d pixel arrays in artificial eyes, and goes pretty deep into explaining why a fractal search is the most efficient, and in particular a fractal with D = 1.5, then gets into what things we find aesthetically pleasing, and it's actually things with fractal dimension that match the eye's natural fractal search dimension. Quite interesting overall.

[Tels]

Thanx for the link, love to read such stuff

 

Wether a simple eyeball movement simulation would be better than none is something to argue for, tho

[demagogue]

I studied cognitive science in undergrad, so of course this stuff is always interesting to me. This is focusing on some stuff and leaving some other stuff out, but really interesting that the search algorithm would match an equation so regularly. (As one prof would say, any time you see a graph that has some regular curve you can see, or matching some equation it's time to break out some champagne because it's so rare and a big deal.)

 

Two things I studied I'd be interested to see how it connects. One was the two-way street between top-down and bottom-up processing with visual scene construction (part of the search processing)... Bottom-up is what you expect of course, as low level features (lines & surfaces) construct or push up what an object or situation is, then you interact with the object (i.e., it sees certain fractal patterns to help recognize what the object or situation is). But what was interesting was how much top-down pushing too; when higher level processing recognizes what an object or situation is can modify perception of even the lowest level stuff & basic features (the lines and surfaces; and there are pathways from high level areas directly to the nuclei right behind the retina modifying signals at the lowest level, that was the surprising part, i.e., it recognizes the object then that tells it how to shape or track its silhouette & features). But really interesting how tangled the hierarchy of processing is, top-down and bottom-up going on in tandem. And recognizing fractal paths also fits with that kind of idea (since fractal geometry itself has an interesting bottom-up and top-down interplay.).

 

The other thing was the role of utility functions in visual search. Might be tangential to what this study was focusing on. But just to note, IMO one of the most interesting studies in vision cognition recently was Glimcher's studies on saccade planning (how the eye flits around) in the LIP area... where they found a literal economic computation of the relative expected utility function of looking at particular features over others. They found a literal computation of Nash's equilibrium theorem, which AFAIK (according to his book anyway) it's the first time they've found Nash's theorem literally computed in the brain. So with this fractal study, I'd be interested in how it might fit with that set up & the economic story behind it, if it optimizes the utility of searching versus the costs in time & opportunity costs. They spoke a little to that, but just be interested in reading more about it, and how that might fit with the optimization & economic architecture of visual search generally (Glimcher's & David Marr's program).

 

All around interesting stuff.