U of S | Mailing List Archive | alt-photo-process-l | Re: OT: 16 bit editing myth or reality?

Re: OT: 16 bit editing myth or reality?



Yves,

I understand you like statistics, but the scope of your
problem is entirely deterministic other than you assumed the
input signal to have normal distribution yesterday. Your
manipulation, quantization and evaluation of the mean square
error are completely deterministic.

I do not agree about the normal distribution
assumption. Almost no image I edited had anything like normal
distribution. Tails are usually much much thicker than
normal. Many of my images look more closer to uniform than
normal, with multiple local peaks, skewness, etc. that are not
a feature of normal distribution. Normal distribution is seen
in many places in nature where the observed quantity is an
aggregate results of many small factors. However, image pixels
in usual pictorial photography are not this type of statistic.

I also do not agree that mean square error is the best measure
for the error for your study.

From: Yves Gauvreau <gauvreau-yves@cgocable.ca>
Subject: Re: OT: 16 bit editing myth or reality?
Date: Mon, 17 Dec 2007 06:15:27 -0500

> Ideally, this would need to be done on more then one image,
> 30 or more would give us what is called a confidence
> interval that the difference is or is not significant based
> on the (variant) CIE dE method I used.

No, you wouldn't need to do that. Since everything is
deterministic, all you need is a reasonable assumption for the
density function for the input data. You can make it up or
empirically obtain this using any of the density estimation
techniques. Then the error measure can be obtained numerically
(or analytically, if you keep things simple) with
deterministic certainty.

Now I like to add that the results obtained above will be
sensitive to the choice of numerical manipulation to the
image, quantization method and presence of small noise added
to the input.

Furthermore, the issue becomes more complicated when spatial
dependency of the input statistic as well as the manipulation
is considered. Such is the reality in image processing. If
your debate was 12-bit vs. 16-bit, the conclusion may be a
matter of judgement. But between 8 and 15 or 16 bits, I think
the practical decision is a no brainer.

--
Ryuji Suzuki
"I can't believe I'm here.
People always say that I'm a long way from normal."
(Bob Dylan, Normal, Illinois, 13 February 1999)