What you've described is a smoothing algorithm (that operates in
colorspace), which basically takes you down a notch in terms on contrast and
sharpness. Think about how to deal with the areas where you DO want a
quick transition between colors, like at the edge of a leaf- or where a
green mountain meets the sky. With the algorithm you've describe below, the
edge would be softer and contain less information than if left alone. (I
mean, we're all shooting Provia 100F, for a reason, right? :-)
Or, what about me; I take a lot of astrophotos, and I want my stars to be
high frequency and pinprick sharp. There's no "logical" reason why a
0xffffff pixel should be next to a 0x000000 pixel, but it's what I want!
And actually, in most of the systems we're dealing with, it's 2^24 = 16.8M
colors or even more with the new/higher end scanners.
Fire up an old copy of qbasic or something; this stuff is pretty easy to
mess around with.
[mailto:email@example.com]On Behalf Of Lynn Allen
Sent: Sunday, April 29, 2001 10:36 AM
Subject: Re: filmscanners: Noise correction algorithms
>How does a computer know that the blue in your picture is from the sky? Or
that the red in your picture is a sunset and not a sportscar? :-)
I'll give a serious answer to a humorous question--if I may deviate from my
usual pathways. ;-)
I believe it has to do with matrixes, which I don't pretend to understand,
on the programming level. If a surrounding field is "blue" and one or
several "red" pixels show up in it, are they supposed to be there? Logic
would say, "No." So the next question the algorithm should ask is: "What
color *should* this pixel be?" Since it has no logical reference except the
adjoining colors, it will have to answer itself, "Well, it should probably
be the same blue as those pixels on either side of it--if it weren't, this
Yo-yo who thinks he's controling this program wouldn't be asking for me to
do this, would he?"
The same would apply to sunsets, although there *is* red in there. The
program should say, "OK, there's red in here, but that pixel is too *dam*
red! Let's change it to a proper color."
There are 256 colors for that pixel to be. One of them has to be nearly
right. I don't ask for perfection from my computer, because that, of course,
is reserved for *me*! ;-) All I ask of it is, "Just work with me here, OK?"
FREE! The World's Best Email Address @email.com
Reserve your name now at http://www.email.com