Re: [css3-color] ICC implementation

Chris Murphy wrote:
> On Jul 12, 2008, at 7:12 PM, David Woolley wrote:
>>
>> Example of images that are clearly gamma 1.0, and have been for 
>> several years, in spite of my pointing out the problem, and their 
>> being obviously wrong, supplied off list.
> 
> I think you're confused about how tone reproduction curves work. A 
> digital raw capture from a camera is linear encoded. You can think of 
> each sensor in the chip as a photon counter. Its response is linear 
> which means it captures a scene with double the luminance, the sensors 
> will report double the values; if the scene triples in luminance, the 
> sensors report triple the values, etc. It's not logarithmic like human 
> vision, or any of our displays.

Agreed.  But...
> 
> Absolute colorimetry of any scene rapidly gets very bright, so most 
> properly exposed captures to a camera do the same thing, and will be 
> represented substantially more by the bits above 50% than below it. 

Not true.  In the simple case of a CRT display, the output is 
approximately proportional to the 2.2th power of the input, which means 
that a 50% absolute brightness on the display corresponds to a 72% input 
value, or a 50% input value produces only a 21% output value, i.e. if 
you feed a linear image into a CRT (sRGB colour space) it will look dark.

> There are many more bits dedicated to the top/lighter 1/4 than the 

Only after applying gamma correction, but we are talking about 
uncorrected images.

> bottom/darker 3/4's because of this. Human vision clamps down on this to 
> keep us from having sensory overload. Turn on a second light with exact 
> equivalence to the first and the room does not get twice as bright, as 
> far as human vision is concerned. To the camera it would be.

True, although I know it as being that subjective brightness steps are 
less at high brightness, although with a gamma of rather less than 2.2. 
  I believe that the Mac uses a gamma law that approximates this 
behaviour in the human eye, so requires less corrected images than does 
sRGB.

However, this is a red herring in terms of making the display brightness 
match the scene brightness, as the same eye companding curve is used in 
both cases, and one can't directly measure the brain's brightness value, 
and it wouldn't be useful to do so.

The equal perceptual steps issue is relevant, if one is using a 
relatively low resolution digital coding, which is why the Mac doesn't 
use linear gamma, even though that is better for image manipulation.

> 
> If you were to take this raw linear capture, not tone map it, and put it 
> up on a display, it will look terribly bright and washed out. That's 
> true if it is a correctly exposed shot.

As explained above, about 72% of the display transfer function is below 
50% absolute brightness.

> 
> This is why no manufacturer does what you are claiming they do, which is 
> they have images that have not been tone mapped, and are still linear. 
> You have it exactly backwards.

No you do.  This does confirm my point that very few people actually 
understand gamma.
> 
> The images in your example are too dark. If lightened in Photoshop, 
> shadow detail becomes obviously noisy, which is a hallmark indication of 
> underexposure with today's digital capture technology.

The give away that they have a gamma problem is that adjusting the white 
level produces bad results, but adjusting the mid-level (the dumbed down 
name for gamma in the tone correction dialogue on most image processing 
programs) does produce much better results.  I seem to remember that 
there is also some element of underexposure in these images aa well - 
that should be easy to see on the histogram display, as there will be a 
gap at the white end.

With the Wikipedia image, I did, at first, assume underexposure, but 
correcting white level was obviously wrong, so I think did a gamma 
correction, and found that applying a correction of 2.2 (the mid-tone 
sliders are generally calibrated in gamma, even though they don't 
document this) looked right to me.
> 
> So I think you're confused on the difference of an image that is a 
> linear encoded, and an underexposed image.

Another good way of seeing that linear images dark displays is to use a 
scanner which you know has not had its gamma correction table loaded 
with software that defaults to no correction, e.g. in my case, a Plustek 
9630P using sane, under Linux (I've seen the driver source code and I 
know that it doesn't use the scanner gamma correction capabilities). 
The 30 in the model number means that it uses a 10 bit A/D, but that 
only helps if you load the gamma table.

A more subjective test is to text anti-aliasing on a black and white 
image using a typical image processing program.  One has to lighten the 
image to get the best results (anyone doing this on half tone or doing 
alpha blending needs to darken sRGB, transform, then lighten it to get 
good results.  On your theory you would have to perform the reverse tone 
corrections.
-- 
David Woolley
Emails are not formal business letters, whatever businesses may want.
RFC1855 says there should be an address here, but, in a world of spam,
that is no longer good advice, as archive address hiding may not work.

Received on Saturday, 9 August 2008 21:02:19 UTC