Re: [Openexr-devel] f-stops and steps. How are they related?

[Yves Poissant]

Hi Florian,

I just finished reading Charles Pontyon and Geg Ward documents you pointed 
me to. I want to thank you for your patience.

I now understand why my reasoning about breaking down the f-stops segments 
into equal number of levels was fundamentally wrong and why my assumption 
that the encoding could be thought as a floating point was also off.

Actually, I now see that the traditional 8 bit encoding is similar to the 
lowest non-normalized portion of the EXR format. That is, in linear 
encoding, each higher f-stop gains in precision and the lowest f-stop 
segment have very little precision. I used to know about Gamma adjustment 
pretty well but failed to recognize the implication on the resolution. The 
"High Dynamic Range Image Encoding" paper by Greg Ward made it all fall in 
place.

By using a binary floating point representation, we have a representation 
that keeps a very good resolution, and each f-stop segment is exactly 
represented like the other ones, thanks to this perfect match between the 
inherent power of two scaling of the binary exponent and the power of two 
nature of f-stops.

As a side note, concerning my monitor gamma adjustment, I adjusted the gamma 
(which is set to 1.65 on the video card driver) so that the set of gamma 
patterns that are available on the Net (and I tested all of which I could 
find) does look OK. So in the end, I don't really know the exact total gamma 
I have since the monitor itself, a flat Sony trinitron, already have a 
pretty good hardware gamma correction when I compare it to other monitors 
around. You are correct that adjusting my monitor instead of the data does 
produce contouring in dark portions of images. This is most noticeable with 
strongly compressed jpeg images (although I only produce jpeg images as 
final output) and a little lesss noticeable in non compressed images. But 
that is the price to pay for wanting to work with linear data. I used to do 
a lot of image manipulation, especially RAW digital photo data (which is not 
really linear but that is another story) and needed to work in linear 
encoding for all those operations.

Regards,
Yves Poissant
Hash Inc.

----- Original Message ----- 
From: "Florian Kainz" <address@hidden>
To: "Yves Poissant" <address@hidden>
Cc: <address@hidden>
Sent: Thursday, January 06, 2005 1:32 PM
Subject: Re: [Openexr-devel] f-stops and steps. How are they related?


I don't know your setup or what you are doing, but if your monitor
is calibrated, then it most likely has a gamma somewhere between
2.2 and 2.5.
The way typical computer and video hardware is set up, the voltage
of the analog video signal produced by your computer is proportional
to the 8-bit pixel value you store in the computer's frame buffer,
but the amount of light emitted by your monitor is not proportional
to the voltage of the video signal.  This means that if you load the
file's pixels directly into the computer's frame buffer, the pixel
data are not "linear", i.e. the pixel values are not proportional
to the amount of light output by the monitor.  With 8-bit pixels,
linear pixel data are not even desirable; they lead to severe
contouring artifacts in the dark regions of an image.
Again, for more details, I recommend Charles Poynton's Gamma FAQ.
For even more details, I recommend Poynton's books on video
technology.

Florian