Does changing exposure (camera or pp) produce hue shifts? (ETTR and HDR related)

[Tim Lookingbill]

Very interesting and challenging analysis, methodology and work, Guillermo.

Was wondering if someone was going to tackle this subject though I was hoping you would've attempted to match pp exposure/luminance changes to the behavior of changing luminance as it actually appears in a real scene instead of focusing on how a digital camera records it through exposure which seems to be a black box of standards based assumptions and guesses designed to deliver a reasonable facsimile reconstructed on a computer and display preview.

The camera is just a dumb computer that hands off linear sensor data to software that does most of the work of interpreting what was captured based on those established spectral assumptions.

It's just that I feel photographers expect software sliders and other editing tools should act on the preview when editing such data to mimic some behavior whether real light or shutter mechanism controlled light that hasn't been that well defined yet if not based on old film exposure standards. Has any digital imaging programmer ever looked at how real light acts on color and make notes about it to base algorithms on rather than rely on decades old color models?

IOW why not just create algorithms that make the software sliders imitate real light changes and how it actually affects color as viewed in a real scene? Skip the middle man, the camera's sensor electronics. It's all reconstructed anyway through software.

Anyway, good work on this. Very informative.

[Guillermo Luijk]

The scope of the test was indeed very narrow: just to find out if the theory about how a digital sensor works (basically a linear photon counter), holds true when changes in camera exposure are compensated by software exposure. But the consequences are important because they allow to completely dissociate camera exposure from final image exposure without any issue or side effects regarding colour rendition, something that didn't happen with film.

And the simplest and more precise point at which software should perform exposure correction to 100% mimic camera exposure is at the very beginning, when RAW data are still linear and colour is nothing else but a RAW {RGB} ratio.

If a given camera exposure T=0,5 s produces RAW numbers:
R=400
G=700
B=300

Doubling exposure to T=1 s will produce (I'm ignoring Canon black point offset for simplicity):
R=800
G=1400
B=600

Potential captured colour is the same because RAW {RGB} ratios are the same. Now if the user sets the RAW developer slider to -1,0EV, the software just needs to divide by 2 all those RAW numbers getting virtually the same RGB data as we would have obtained with T=0,5 s and the slider set to 0,0EV:
R=800/2=400
G=1400/2=700
B=600/2=300

In the real world other non colour-related effects take place: the T=1 s shot will have improved SNR over the T=0,5 s and in exchange could clip up to 1 extra stop of highlights information. But regarding colour (RAW {RGB} ratios in non-clipped areas), both RAW files are potentially clones while exposure correction is carried out at the right place. Why commercial software usually don't do it exactly in this way, allowing (slight) colour shifts when touching the exposure slider, falls out of my understanding. This test proves that a genuine Exposure slider is possible.

Regards

[Jim Kasson]

The scope of the test was indeed very narrow: just to find out if the theory about how a digital sensor works (basically a linear photon counter), holds true when changes in camera exposure are compensated by software exposure. But the consequences are important because they allow to completely dissociate camera exposure from final image exposure without any issue or side effects regarding colour rendition, something that didn't happen with film.

<snip>

...regarding colour (RAW {RGB} ratios in non-clipped areas), both RAW files are potentially clones while exposure correction is carried out at the right place. Why commercial software usually don't do it exactly in this way, allowing (slight) colour shifts when touching the exposure slider, falls out of my understanding. This test proves that a genuine Exposure slider is possible.

Excellent work, as usual. Congratulations.

Here's a more quantitative look at the effect of Lr PV 2012's Exposure control on chromaticity than your excellent Macbeth chart images:

http://blog.kasson.com/?p=2994

By the way, I believe you are using the world "hue" to mean what's left of a 3D color representation once luminance is removed. If that is indeed the case, the standard term in color science is "chromaticity", measured as a two dimensional quantity in xy, u'v', or horizontal cuts through Lab or Luv.

Thanks for all your work here, and in the past.

Jim