Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uhog.mit.edu!europa.eng.gtefsd.com!howland.reston.ans.net!EU.net!sunic!news.funet.fi!cs.joensuu.fi!jlavi
From: jlavi@cs.joensuu.fi (Jarkko Lavinen)
Subject: Re: Are 24-bit video adapters toys?
Message-ID: <1994Oct25.113957.13989@cs.joensuu.fi>
Organization: University of Joensuu
References: <1994Oct13.121017.481@cs.joensuu.fi> <37m17b$mh6@kodak.rdcs.Kodak.COM> <usenet-1410941051060001@lowry.eche.ualberta.ca> <37uo35$2vb@kodak.rdcs.Kodak.COM> <usenet-1910941241490001@lowry.eche.ualberta.ca> <388gj0$k0c@kodak.rdcs.Kodak.COM>
Date: Tue, 25 Oct 1994 11:39:57 GMT
Lines: 103

bryant@neural.Kodak.COM (Steve Bryant) writes:

>In fact, my colleague here says that he has read research articles stating
>that the eye is not capable of resolving less than 3 parts in 256. Granted,
>this is pure hear-say evidence and I dont have the reference to the article,

Once again the problem is that this doesn't say anything about 
the density difference. One should say the result rather in 
ratios or percentages, something, something like that human 
eye can't see a 1% difference in luminance or whatever. Ratios 
can be easily converted to density difference by taking a 
logarithm in base 10.

Earlier Alan Roberts wrote:

>to exceed 50:1. To view a picture with contrast of 1000:1 you will
>need a very good (low flare) CRT and a completely blacked out room.
>I have measured, in a lab, contrast on a CRT of over 3000:1, but that
...
>then you will be very lucky to get contrast of more than 35:1. This
>results from flare in the glass of the CRT, and in the AGC action of
..,
>Basically, you might be able to store pictures with contrast of 1000:1
>using 10-bit values, but you'll never be able to see them properly on
>a CRT.

One could perhaps have a flare model of the CRT and use that then during
the viewing to compensate the loss of contrast. The viewing program
would first estimate whether it can view the requested contrast or not
and then try to make best out of the situation. This approach could
also easily compensate dark corners of the CRT.

To get more color accuracy, one cheap approach might be to simply 
dither 30-bit images to 24-bit. When viewing photographic continous 
tone images this dithering disappears in high resolution and would 
mimic 30-bit Truecolor. Similar approaches are used to simulate 24-bit 
Truecolor in non-24-bit adapters but here the effect would be much more
subtle because 24 is already almost continous. When mimicing 24-bit in
16-bit the effect is much more coarse.

Neverthless this would requite that density reference to be coded 
somehow in image. I think PhotoCD does this because luminance is encoded 
in logarithmic steps. For my purposes PhotoCD takes too much space
(7000 images would fit only as a thumbnail in CD) and I would rather 
do with JPEG-like approach, but plain JPEG doesn't store density 
information. Perhaps JPEG could be modified to support 10 bits and 
contain density reference. Otherwise one doesn't know what gamma 
correction the data contains. One may also ask that whats the use 
of talking about lossy image compression with 30-bit accuracy... :-(

Earlier Ron Natalie wrote:

>I have some test images and we've never managed to get down to
>being able to see 3 step differences across a wide range of
>intensities.

I made an experiment with using netpgm package and xv 3.0,
with following command:
	pgmramp -lr 1024 100 | xv -owncmap -

The monitor I had was IBM 6091 19i, hooked to IBM RS/6000 with
the low-end GT1 adapter on 1280x1024/256 mode. The room was as
dark as it can be - it was night, windows were shaded, lights
turned off etc.

The pgmramp produces a continous linear scale 0..255 with given
geometry. I rotated the scale vertical, then Maxpected, then
zoomed into various parts to get the bars more visible. I then varied
brightness with maximum contrast and tried to see boundaries
of shades. Both in the dark and bright end there were areas
where I was able to see fringes with difficulty but not beyond.
In between these ends there were a scale where steps were
visible without difficulty when zoomed enough.

I list my results as 4 numbers: First one is the point where
I can first see steps. Last one is where steps disappear.
Second and third describe the scale where steps appear
continuously and are seen relatively easily. I tried first
with maximum contrast varying brightness and then with
medium contrast. My results were:

hi  con, low bri: 50, 60..180, 220
hi  con, med bri: 30, 30..130, 190
hi  con, low bri: 15, 20..130, 150
med con, med bri: 50, 65..140, 190

Needless to say that these numbers are very subjective. One 
should actually draw a sensitometric curve like those used
for photographic media. Also I should have done this in all
three guns separately etc.

It would be interesting to hear what kind results people get
from high-end machines like Silicon Graphics with this
setup I used. In a way these numbers are pretty meaningless
without sensitometric control, but this is a quick and dirty
experiment.

According to the Gamma Measurement Image in Monitor Gamma page at:
	http://www.cs.cmu.edu:8001/afs/cs/user/rwb/www/gamma.html
the gamma on the 6091 was 2.2 (max contrast).

	Jarkko Lavinen
jlavi@cs.joensuu.fi, http://balrog.joensuu.fi/~jlavi
