Re: dithering by optimizing perceptually-weighted error in the Gabor transform

As I learned it a couple of decades ago, most neural responses,  
including vision, can be modeled by a "mexican hat" response, which  
is what your difference-of-gaussians approximates.

 From the art side, we know that it's possible to acceptably render  
scenes with a much reduced dynamic range by paying attention to the  
edge qualities: hard, soft, or lost.[0]  Human vision isn't so much a  
leaky integrator as a leaky differentiator, which also explains why  
placing specular highlights ("catchlights") in eyeballs is worth so  
much for portraits, whether photographic, painted, or even sculpted.   
Contemporary broadcast TV also seems to make heavy use of theatrical  
three-point lighting to improve the perception of the rendition.

At this point my guess as how to best improve your engraver would be  
to somehow arrange a cutover so that high-amplitude edges are  
strengthened while low-amplitude edges are weakened or even lost,  
which should eliminate quite a few of the compressive artifacts.  Not  
sure if this could be done locally; my impression is that you'd  
somehow want to arrange for appropriate gradients matching up the  
values in the spaces between the hard edges as an automated  
approximation to what airbrush artists[1] accomplished by spraying  
smooth gradients in between frisketed edges.

-Dave

[0] cf http://www.xent.com/pipermail/fork/Week-of- 
Mon-20101122/058148.html
[1] technical illustration may also be a useful source of ideas

-- 
To unsubscribe: http://lists.canonical.org/mailman/listinfo/kragen-discuss