Quick and dirty comparison of color reproduction

[Erik Kaffehr]

I was playing around with Matlab and comparing color interpretation between different cameras:

Comparing GFX 100 with Sony A7rIV, convertion in Lightroom using color profiles generated by Lumariver. Differences are small. Color difference is Delta E 2000, with L channel normalised.

GFX 100, comparing processing in LR with Lumariver profilles and Capture One. Here we can see that Capture renders colors quite a bit different from Luamriver on the GFX 100 sensor.

This compares GFX50II with X1D, both use the same Sony sensor but CFA (and other toppings) may be different. Lumariver profiles yield very similar colors.

P45+ compared to GFX 100, using Lumariver profiles. Some colors of the ColorChecker render a bit differently between GFX 100 and P45+.

This comparison is not Jim Kasson kind of solid comparison, more like Quick and Dirty comparison.

My 'take away' is that:

• Color profiles may matter more than sensor differences.
• Fuji GFX, Hasselblad X1 and Sony A7rIV may deliver pretty similar color when using LumaRiver generated profiles.
• The P45+ I used yields different reproduction, even when using LumaRiver. So sensor design may play a role even when using the same color profiling tool.

As stated before, this is not solid engineering quality work. But it may be food for thought...

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic tends to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[tex]

• Color profiles may matter more than sensor differences.

For certain kinds of work, they make a huge difference! But I'd say that all of that work is in the pro category---mostly product photography and its related niches (like cultural heritage work).

For personal artistic work, does color fidelity matter at all? I would argue no.

 

[Erik Kaffehr]

• Color profiles may matter more than sensor differences.

For certain kinds of work, they make a huge difference! But I'd say that all of that work is in the pro category---mostly product photography and its related niches (like cultural heritage work).

For personal artistic work, does color fidelity matter at all? I would argue no.

Hi Tex,

My posting does not look at accuracy, but looks at similaraties and differences between sensors and profiles.

Accuracy is not involved at all.

Best regards

Erik

 

[tex]

So, now I'm not understanding the point so much.

 

[Erik Kaffehr]

A short explanation...

This plots shows similarity between two sensors. The two sensors would yield almost indistinguishable results usining the same profiling methodology.

Comparing the GFX data above with the ColorChecker reference shows how accurate the data is. JND (Just Notiacble Difference) on the Delta E2000 comparison is intended to be one, so many of the errors would be noticable.

Usiing Capture One, here with an older Phase One back probably using a Kodak designed CFA shows less accuracy, due to both the CFA and hot mirror design by Phase One, but also because C1 adds it's own look.

Here are the data:

GFX100 vs A7rIV in LR DE 2000 (corr), avg = 0.82 std = 0.50 median = 0.80 max = 2.01
ColorChecker vs GFX100 in LR DE 2000 (corr), avg = 2.50 std = 1.31 median = 2.61 max = 4.75
ColorChecker vs P45+ in C1 DE 2000 (corr), avg = 3.27 std = 1.80 median = 3.71 max = 6.37

I tried to remove the effect of the tone curve in this comparisons, by normalizing the L channel according to the grey samples of the color checker.

The itention with the original posting was to illustrate that different sensors may yield near identical results when profiled in the same way.

As a side note, it is probably best to start with accurate color and adjust to taste than to start with less accurate colors and adjust to taste, that because adjustments can cause side effects.

Another note is that we need to have correct white balance.

Best regards

Erik



--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic tends to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Gandolphi]

My feelings exactly. Studio work, Profesional work where a certain brand colour needs to be spot on (New car colour for the year), yes but for most things its not so important.

• Color profiles may matter more than sensor differences.

For certain kinds of work, they make a huge difference! But I'd say that all of that work is in the pro category---mostly product photography and its related niches (like cultural heritage work).

For personal artistic work, does color fidelity matter at all? I would argue no.

 

[Stan Disbrow]

Hi,

There is so much at play between cameras that I don't pay this much mind. I just get something new and go shoot something, anything, and do the same with what was old. Same place and almost the same time. And then process with the same software and see what I think.

The only camera I had which was far away different when it came to color was the Kodak DCS 620x. And that used a CYM dye set for the CFA specifically to let more light thru and have a cleaner higher ISO.

The later 720x was the same thing only different in that the images weren't nearly as yellow tinted. And this was processing both thru Kodak's own software.

Stan

 

[just Tony]

The number of available settings in LumaRiver is astonishing, but I’ve settled on one “middle of the road” recipe for the default profiles of all of my cameras which LightRoom then uses specifically to each. That gives me a starting point for my creative edits that at least feels identical among my Nikons, and the GFX 100S isn’t far off from them at all, to the point where I have no concerns in practice.

A close side by side comparison of Color Checker patches will reveal something subtle - and that’s in the calibration set which in a perfect world would have zero error. But we all know that the passbands of the CFE aren’t Gaussian. Those little bumps and dips must have some effect that we can’t completely dial out.

In landscape scenarios there are larger variables anyway. Light bounces off object A which isn’t neutral gray, onto object B. Color rendition isn’t on my list of how I choose a camera for a task.

--
Wag more; bark less.

 

[Erik Kaffehr]

The number of available settings in LumaRiver is astonishing, but I’ve settled on one “middle of the road” recipe for the default profiles of all of my cameras which LightRoom then uses specifically to each. That gives me a starting point for my creative edits that at least feels identical among my Nikons, and the GFX 100S isn’t far off from them at all, to the point where I have no concerns in practice.

A close side by side comparison of Color Checker patches will reveal something subtle - and that’s in the calibration set which in a perfect world would have zero error. But we all know that the passbands of the CFE aren’t Gaussian. Those little bumps and dips must have some effect that we can’t completely dial out.

In landscape scenarios there are larger variables anyway. Light bounces off object A which isn’t neutral gray, onto object B. Color rendition isn’t on my list of how I choose a camera for a task.

I wouldn't expect a perfect match, as there are 18 color patches on the ColorChecker and we only have coefficients in the color conversion matrix, so we do need an optimization for chosing those 6 coefficients, that may be the reason it is sometimes called 'the compromise matrix'.

We can obviously have a LUT matching all available color patches exactly to reference.

Just my thinking, of course!

Best regards

Erik

 

[JimKasson]

I wouldn't expect a perfect match, as there are 18 color patches on the ColorChecker and we only have coefficients in the color conversion matrix, so we do need an optimization for chosing those 6 coefficients, that may be the reason it is sometimes called 'the compromise matrix'.

We can obviously have a LUT matching all available color patches exactly to reference.

For one illuminant.

 

[Erik Kaffehr]

I wouldn't expect a perfect match, as there are 18 color patches on the ColorChecker and we only have coefficients in the color conversion matrix, so we do need an optimization for chosing those 6 coefficients, that may be the reason it is sometimes called 'the compromise matrix'.

We can obviously have a LUT matching all available color patches exactly to reference.

For one illuminant.

Thanks for that remark.

Best regards

Erik