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

[Brian Willoughby]

Hello all,

I hope the list members do not mind if I reply to Yves' questions, even  
thought the discussion is not purely related to EXR.


Considerations of gamma can get quite complicated when you consider all the  
input devices, standards for file formats and image transfers, and all the  
output devices.  For the sake of simplicity, I find it easiest to break it down 
 
to two elements: a computer display buffer and a CRT.  Understanding that  
relationship allows extrapolation to the other elements.

First of all, I think it is important to understand why we work with a linear  
representation of light levels.  The reason is that most processing is  
mathematical, and the math only works correctly with linear data.  The same  
functions are applied to every pixel, regardless of light level, so therefore  
the mapping from perceived light level to numeric code should be linear.

Next, it is important to understand how a CRT works.  The digital codes  
(usually 8 bits per component) are sent to DACs (Digital to Analog Converters)  
which convert to voltage.  In a CRT, these voltages drive electron guns aimed  
at phosphors.  The issue is that phosphor does not react linearly.  When given  
a voltage that represents light levels in a linear mapping, the phosphor does  
not produce a linear light output.  The curve that relates input voltage to  
output light level is exponential.  Gamma, at least for a CRT, is a means to  
correct for this exponential curve.  I believe that most phosphor responds with 
 
a gamma of 2.2, which is why you see that value recommended.


After you understand the basics, then you have to consider that not all  
displays are CRTs, especially not now.  New display cards have digital output,  
so the monitor itself must have the DACs and must apply gamma if needed.  Plus, 
 
there has never been a universal standard for gamma, so many people who  
produce computer images will put a gamma curve into their image (hopefully  
*after* all processing is finished for the image).  Then, you also have the  
issue of non-linearity from input devices, unless you're dealing with purely  
rendered images.

I tend to scan without gamma, but the images look really dark on the average  
monitor, unless the display card and monitor are set to 2.2 gamma.  If I scan  
with the default gamma (which is 2.2 on my scanner), then images are blown out  
if I also set my system to 2.2 on output.

Remember that you may be dealing with images that are truly linear, or you may  
be working with images that already have gamma applied so they will look good  
on the average display attached to the average operating system.  If the image  
has less than 2.2 gamma, and you're viewing it on a CRT, you'll have to apply a 
 
gamma of less than 2.2 to make up the difference.  Applying 2.2 gamma (with  
the display card and monitor settings) to image data that already has some  
gamma will be too bright.

There have been standards proposed for TIFF to document the embedded gamma  
with a tag.  I'm not sure how widespread this is.  Also, many advanced  
operating systems have API that will gamma correct individual windows or even  
individual images within a window for gamma.  Apple had ColorSync.  The OmniWeb 
 
web browser will gamma correct each image on a web page based upon information  
found in the image file.  If any of these technologies are manipulating your  
image on the way to the framebuffer and monitor, than gamma correction can  
easily get out of control.

P.S.  One thing to be aware of is that many display cards only have 8-bit  
DACs.  If your framebuffer is also 8-bit, and the display card is mapping the  
gamma, then you will have lost some codes.  In other words, you will have fewer 
 
than 256 steps of voltage.  Better display cards have a 10-bit DAC, and the  
RAMDAC lookup tables are 8-bit input, 10-bit output, so you will retain 256  
unique steps even after conversion to the exponential gamma curve.

Brian Willoughby
Sound Consulting