436 replies to this topic

[BQ Octantis]

Where are the colorchecker patches in that plot??

Paint. Specifically, color-matching cards.

[timaras]

The sun came out today after several days and I repeated the work in post 388 to capture the colorchecker using a measured spectrum, this time daylight sun (it was around 1pm, with the sun at 60deg altitude), instead of the full spectrum 4400K bulb. The goal was to see how much the choice of illuminant when producing the calibration (forward) matrix affects the color error.

 

Some interesting (and unexplained) findings: i) errors were overall higher compared to calibrating with the 4400K bulb, and ii) it was progressively worse as CCT went up.

 

I used an optical fiber pointed at the sun to capture the spectrum while at the same time taking a photo for calibration. The spectrometer was managed by a battery-powered miniPC.

 

 

Below is the captured spectrum (green line) compared to other standard illuminants and the bulb. Equivalent CCT of the sunlight spectrum was 5400K.

 

 

And these are the color errors presented by dcamprof when creating the profile of the Canon R6 camera, while selecting different illuminants:

 

Indicated Illuminant    ΔΕ (av)    
StdA                    2.46
D50                     2.67
5400K                   2.94
Sunlight 5400K (actual) 2.87
D55                     3.11
D65                     2.96
D75                     4.46

 

It's worth noting that the estimated WB gains (which is the most important factor in calibration) calculated by dcamprof when feeding the 5400K measured spectrum, were very close (within 5%) to the WB gains provided by Canon for 5500K blackbody illuminant.

Edited by timaras, 14 June 2025 - 05:18 PM.

[FrankieT]

Some interesting (and unexplained) findings: i) errors were overall higher compared to calibrating with the 4400K bulb, and ii) it was progressively worse as CCT went up.

Perhaps upload the raw colour checker image with the measured illuminant spectrum file somewhere and I can try to reproduce the results.
 

[timaras]

Perhaps upload the raw colour checker image with the measured illuminant spectrum file somewhere and I can try to reproduce the results.
 

Here is a folder with the files, I included both the sunlight and the previous full spectrum "biolight" data.

 

(there is also a json version of the spectra normalized to 100% at 526nm)

[BQ Octantis]

Here is a folder with the files, I included both the sunlight and the previous full spectrum "biolight" data.

 

(there is also a json version of the spectra normalized to 100% at 526nm)

 

Since you've got a way to measure spectrum, it seems silly to me to rely on matching methods and tools that presume a reflectance spectrum for the target swatches. Why not measure the spectrum directly off each patch (in addition to the illuminant)? Then you can separate the printing error from the camera measurement error…and you'd advance the state of knowledge beyond what's published…

Edited by BQ Octantis, 15 June 2025 - 05:40 AM.

[timaras]

Since you've got a way to measure spectrum, it seems silly to me to rely on matching methods and tools that presume a reflectance spectrum for the target swatches. Why not measure the spectrum directly off each patch (in addition to the illuminant)? Then you can separate the printing error from the camera measurement error…and you'd advance the state of knowledge beyond what's published…

I could measure directly, but it is quite time consuming with little further benefit for astro applications... I am happy with the color I obtain. The dependence on illuminants interests me more as setting a color temp and WB have immediate visual effects.

[BQ Octantis]

I could measure directly, but it is quite time consuming with little further benefit for astro applications... I am happy with the color I obtain. The dependence on illuminants interests me more as setting a color temp and WB have immediate visual effects.

If there's anything this thread has revalidated for me, it's that anything worth doing is worth overdoing.

[FrankieT]

Some interesting (and unexplained) findings: i) errors were overall higher compared to calibrating with the 4400K bulb

 

Perhaps upload the raw colour checker image with the measured illuminant spectrum file somewhere and I can try to reproduce the results.

I was able to reproduce your result for the Sunlight spectrum. The target seems underexposed relative to the Biolight image, which may affect the accuracy of the profile. Since I don't have a Canon camera, I compared the Sunlight image to Mark's 600D colour checker referenced in #12, which was also shot under a daylight spectrum. The DCamProf D50 ICC profile for the 600D image has an Av, p90 and Max ΔE of 1.68, 2.59 and 5.43 respectively, which is comparable to the Biolight profile's performance. Thus, I expect a more statistically accurate profile could be attainable for the R6 under a daylight spectrum. That said, whether it would result in a noticeable visual difference is another question.

[timaras]

I was able to reproduce your result for the Sunlight spectrum. The target seems underexposed relative to the Biolight image, which may affect the accuracy of the profile. 

Thanks for looking into this. Is balanced exposure that critical (there is definitely no clipping)? Thankfully I captured multiple exposures, so I have uploaded in the same folder x2 and x4 exposure (1/1250s and 1/640s shutter speed, the first being 1/2500s). I will also check when I get some time this week.

[timaras]

I was able to reproduce your result for the Sunlight spectrum. The target seems underexposed relative to the Biolight image, which may affect the accuracy of the profile. 

In yet another unnecessary exercise, I created profiles using all the captures from that session, at varying exposures (about 3 stops total, at 1/3 increments). Then calculated the color error against the exposure, where as exposure metric I am using the raw CFA values of the colorchecker grey patches. It looks like the exposure is not the problem (note the log scale):

 

 

(If this plot is confusing, just isolate any patch (color line) and look at the error vs. exposure).

 

So there is no better error, and the error remains consistent for a wide range of exposures (2 stops). The raw file I initially provided is the 3rd point from the left.

 

So the jury is still out there on why Biolight colorchecker calibrations create more accurate profiles. 

 

Out of curiosity I also plotted the Canon R6 raw sensor values for each grey patches as exposure was increased. You can see the clipping on the black patch at low exposures and on the white patch at high exposures.

 

Edited by timaras, 18 June 2025 - 04:12 PM.

[timaras]

I wanted to park here several reading material that I found useful as we went through this process in this thread in the last few weeks, for the future generations of readers:

 

Discussions

DSLR/Mirrorless Consistent Color Processing (this thread)

DSLR Processing - The Missing Matrix (the classic CN thread)

M31 Andromeda Galaxy in natural colour (CN sharkmelley's workflow)

ACR Limitations and Pleiades Nebulosity (DP review post)

Keep close to authentic look of object or create stunning, award winning and over processed photo? (stargazers lounge thread)

 

RAW Development

Color conversion matrices in digital cameras: a tutorial (my favorite reading on the subject)

Understanding the In-Camera Image Processing Pipeline for Computer Vision (another excellent tutorial)

Observations on Adobe Camera Raw for Astro-Image Processing (Mark Shelley website)

Developing a RAW photo file 'by hand' (odelama)

Color matrix calculator (conversion matrices)

Bruce Lindbloom (conversion matrices & theory)

Principles and Practices of Chromatic Adaptation

 

Camera profiling

DCamProf - a digital camera profiling tool (tool tutorial)

Lumariver Profile Designer User Manual

How to Make a Better Custom Camera Input Profile (9degbelow)

How to Make a Camera Input Profile Target Shot (9degbelow)

Spectral Sensitivity Estimation Without a Camera

New color specifications for ColorChecker SG and Classic Charts (reference data)

BabelColor ColorChecker Pages

wide-gamut.com (monitor verification)

Edited by timaras, 25 June 2025 - 02:55 PM.

[FrankieT]

If you still have time to edit, I would suggest a few additions to your reference  list:

 

RAW Development

Bruce Lindbloom

Principles and Practices of Chromatic Adaptation

 

 

Camera profiling

BabelColor ColorChecker Pages