Luminous Landscape Forum
Equipment & Techniques => Medium Format / Film / Digital Backs – and Large Sensor Photography => Topic started by: ErikKaffehr on February 22, 2014, 08:14:16 am
-
Hi,
I just posted a comparison of colour accuracy between my Phase One P45+ and my Sony Alpha SLT99 using both Lightroom 5.3 and Capture One.
The findings were essentially:
- The SLT99 has the most accurate colours using LR5.3
- On the P45+ LR 5.3 is more accurate than Capture One when using Capture One's "film curve" which is default.
- Capture One is more accurate on the P45+ when using the linear curve than LR 5.3
- Capture One is significantly less accurate than LR 5.3 on the SLT 99 with both tested curves.
The only case Capture One was better than LR5.3 was when "linear curve" was used with P45+. In all other combinations LR5.3 was more accurate.
Note, this was a measurement of accuracy and not goodness or pleasantness. All alternatives were decently accurate. The results indicate that neither raw converter is markedly superior in colour rendition, but I would say that LR5.3 with a decent DCP profile for the cameras tested seems a bit more accurate.
The results are here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition?start=3
In the figure below P45+ is on the left (using C1 linear curve) and SLT 99 on the right (using LR5.3). Patches Delta E is above 5 are marked.
(http://echophoto.dnsalias.net/ekr/Articles/ColourAccuracy/BestConversion_large.png)
In the figure below P45+ is on the left (using C1 linear curve) and SLT 99 on the right (using LR5.3). Patches Delta E is above 3 are marked, representing a stricter criteria.
(http://echophoto.dnsalias.net/ekr/Articles/ColourAccuracy/BestConversion_small.png)
DeltaE* values from an IT 8.7 target are shown below, based on 288 fields.
Best regards
Erik
-
Nice target, where did you get it, if I may ask?
-
Hi,
I didn't show the actual target, it is just a normal IT 8.7, I bought it Woolf Faust, http://www.targets.coloraid.de. Very nice guy!
Best regards
Erik
Nice target, where did you get it, if I may ask?
-
I like the IT8 target. The only concern with it is it was not designed for color checking DSLRs, it may be highly affected by the source light. Did you use a very high CRI lamp? The normal color checkers use special pigments not affected by the light.
Maybe the IT8 is good for answering the question of camera gamut mapping more than color accuracy?
-
No,
Electronic flash. But I don't think it was a problem, as all alternatives were pretty accurate. The reason I didn't use the colour checker this time was that I wanted a target that was different from the CC, specially as I was using the ColorChecker for calibration.
The ColourChecker did show larger Delta E, than my latest test. See below.
Best regards
Erik
I like the IT8 target. The only concern with it is it was not designed for color checking DSLRs, it may be highly affected by the source light. Did you use a very high CRI lamp? The normal color checkers use special pigments not affected by the light.
Maybe the IT8 is good for answering the question of camera gamut mapping more than color accuracy?
-
Hi Erik, I know you shoot landscape, but was wondering if you have noticed any difference in how skin tones are reproduced?
-
Hi Sareesh,
Unfortunately, no. I have two issues with skin, I have exactly two portraits shot with my P45+, one of those is showing myself, and that one is discussed here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition?start=6 .
The other issue is that as I am a landscape shooter, I cannot really say what skin looks like.
But, I have found some oddball issues with the Adobe Standard profile on the P45+, that I fixed by generating a DCP profile, see below:
(http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/Screenshots/P45+_AdobeSTD_vs_DNGProfiler_small.png)
So I would say that a profile may be reasonably accurate and still produce outliers. The P45+ with the C1 linear profile has outliers at DeltaE* = 17.2, while LR 5.3 with my home made profile has a Delta E of 7.13 for the same patch. Still, the very same patch on the SLT 99 has a Delta of 1.48.
I had downloaded a two raw files from Nikon D3X and Aptus (of some kind). In those images the D3X had a pinkish skin tone with the Adobe Standard profile, but the exposure also included a ColorChecker,
which enabled me to create a DCP profile that removed the pinkish tone. The D3X still looked flat compared to the Aptus.
The image below shows the outliers:
P45+ with capture one on the left, P45+ with LR5.3 at the center and SLT 99 on the right. The LR5.3 files use DCP profiles generated from a ColorChecker Passport.
Best regards
Erik
Hi Erik, I know you shoot landscape, but was wondering if you have noticed any difference in how skin tones are reproduced?
-
Hi,
Here are two screen dumps of the Nikon D3X and Aptus images I have downloaded from the net. I don't know whom to attribute them, unfortunately.
The second one shows the image with Adobe Standards Profile to the left and my DCP profile to the right.
My understanding is that Adobe DNG Profile Editor just adjust an underlying base profile, so it is no a profile just based on the colour checker.
Best regards
Erik
-
Eric,
Nice work as usual from you. Where does one get the color patch compare tool?
I see that you are using Delta E (CIEDE2000), but why did you choose this parameter which includes differences in luminance, hue and chroma. Exposure error can cause large delta E due to the luminance, but the luminance error component may not be perceptible. The delta C calculation eliminates the luminance component, and may be preferable for most purposes. In my experience with Imatest using the colorchecker chart, I always got the least delta E error using PV2010 with linear settings, but the results appeared flat. Many converters boost chroma leading to increased delta E due to the chroma error component, but most observers find the increased chroma pleasing. It might be advantageous to correct for chroma error as done in Imatest (http://www.imatest.com/guides/modules/colorcheck/). However shifts in hue are not welcome.
Regards,
Bill
-
Where does one get the color patch compare tool?
Hi Bill,
PatchTool a commercial product, available from here (http://www.babelcolor.com/main_level/screen_shots/PatchTool_Compare_view.htm).
Cheers,
Bart
-
Hi Bill,
I checked on Delta C, but Delta C ignores both Luminance and Hue. It would be nice get rid of luminance, as it is very tone curve dependent, but I think hue must be kept in any meaningful comparison.
Best regards
Erik
Ps, good input is always appreciated! Thanks a lot!
Eric,
Nice work as usual from you. Where does one get the color patch compare tool?
I see that you are using Delta E (CIEDE2000), but why did you choose this parameter which includes differences in luminance, hue and chroma. Exposure error can cause large delta E due to the luminance, but the luminance error component may not be perceptible. The delta C calculation eliminates the luminance component, and may be preferable for most purposes. In my experience with Imatest using the colorchecker chart, I always got the least delta E error using PV2010 with linear settings, but the results appeared flat. Many converters boost chroma leading to increased delta E due to the chroma error component, but most observers find the increased chroma pleasing. It might be advantageous to correct for chroma error as done in Imatest (http://www.imatest.com/guides/modules/colorcheck/). However shifts in hue are not welcome.
Regards,
Bill
-
Hi,
Here are two screen dumps of the Nikon D3X and Aptus images I have downloaded from the net. I don't know whom to attribute them, unfortunately.
The second one shows the image with Adobe Standards Profile to the left and my DCP profile to the right.
My understanding is that Adobe DNG Profile Editor just adjust an underlying base profile, so it is no a profile just based on the colour checker.
Best regards
Erik
Erik,
On the top comparison, the background on the Leaf looks warmer so the immediate impression is it is off relative to the nikon. Looking at the color checkers the nikon looks much more saturated especially the second column (Caucasian skin tone, blue). Pulling the RGB colors on the bottom grey row the white balance is red on the nikon, it is neutral on the leaf. I think fixing the nikon whibal based on the color checker would turn the person's skin a magenta hue. That profile is way off.
-
On the second one they both seem to drift to the blue at darker tones. The skin tone looks more pleasant with your correction.
-
Hi,
What I found was essentially:
- The Sony Alpha 99 had more accurate colours than the P45+, but the P45+ came pretty close using the linear curve with Capture One
- With the Sony Alpha 99 LR5.3 is more accurate than Capture One
- Generating a camera specific DCP profile helps LR 5.3 in both cases
- The linear curve is pretty linear on Capture One and that probably helps a bit on Delta E based evaluation
- The LR5.3 conversion with camera specific DCP profile is more accurate
- C1 conversions seem to shift deep blues and green towards extreme saturations
- C1 has some pretty extreme colour shifts in blues and greens
Some observations on the methods used.
The 288 colour pathches cover a significant part of Adobe RGB, with some patches slightly outside Adobe RGB.
The fact that Sony SLT and LR5 yields very good colour reproduction indicates that the experiment setup is decently useful. It is highly improbable that methodical error would not show up on Sony but on P45+. Also, the linear curve on the P45+ produces average Delta E close the Sony.
As a side note, the Sony is regarded to have good colour rendition by several authors, like Tim Parkin and also Iliah Borg. It also has a very high SMI-rating at DxO-mark.
I looked at using other Delta values than Delta E. Delta E uses L (Luminousity), C (Chromacity) and H (Hue) coordinates. Ideally I would like to ignore L (as this is essentially a grayscale component).
Using Delta C or Delta L alone would not give any good information. C is essentially saturation, pretty meaningless without hue angle. H is hue angle (essentially). There can be large variations in hue angle for areas of low saturation without a visible difference. So Delta C or Delta H alone give little useful information.
Best regards
Erik
-
Maybe, maybe not. 1 sample of 2 cameras is not scientific. It is not statistically significant. It is very interesting. I'm sure everyone appreciates the effort.
If your experiment is combined with the raws of many cameras provided by Imaging Resource (they all have a color checker in the shot) on the web, then other testers find the same conclusions, you would have enough to make a definitive statement of ACR vs C1. Probably even enough to get them to improve.
I don't dispute your method or findings, I am just saying it needs verification as well as a broader selection of products.
Maybe several forum members can start looking into the available raws. The work could benefit everyone in photography.
-
Hi,
MFD raw images are not a plenty. I have analysed several, and I have always found LR to have an advantage in accuracy. The problem is really that once you calibrate, you calibrate against the 24 colours of the ColorChecker. The real benefit of this test is that it goes "beyond" the colour checker and analyses 288 other patches.
I have no issue with the colour checker, it is just that it covers a limited of colours - albeit probably well choosen - and that it would often be used as base for calibration.
What my test does prove, sort of, is that MFDB is not in general more accurate than a DSLR. I may of course have a bad sample of the P45+ and a good sample of the SLT99. To sort out that question would need a decent sample of each.
Another point may be that the accuracy I have measured using the linear curve on the P45+ was by any means pretty decent. Using a more film like curve may give "better" tonality although it may be less "accurate".
Best regards
Erik
Maybe, maybe not. 1 sample of 2 cameras is not scientific. It is not statistically significant. It is very interesting. I'm sure everyone appreciates the effort.
If your experiment is combined with the raws of many cameras provided by Imaging Resource (they all have a color checker in the shot) on the web, then other testers find the same conclusions, you would have enough to make a definitive statement of ACR vs C1. Probably even enough to get them to improve.
I don't dispute your method or findings, I am just saying it needs verification as well as a broader selection of products.
Maybe several forum members can start looking into the available raws. The work could benefit everyone in photography.
-
As a side note, the Sony is regarded to have good colour rendition by several authors, like Tim Parkin and also Iliah Borg.
wasn't that about A900/A850 ?
-
wasn't that about A900/A850 ?
Yes, the A900 was Sony's attempt to make a photographer's camera with no frills, no BS. A large part was the choice to give up some ISO noise for color separation on the filters.
-
Hi,
MFD raw images are not a plenty. I have analysed several, and I have always found LR to have an advantage in accuracy. The problem is really that once you calibrate, you calibrate against the 24 colours of the ColorChecker. The real benefit of this test is that it goes "beyond" the colour checker and analyses 288 other patches.
I have no issue with the colour checker, it is just that it covers a limited of colours - albeit probably well choosen - and that it would often be used as base for calibration.
What my test does prove, sort of, is that MFDB is not in general more accurate than a DSLR. I may of course have a bad sample of the P45+ and a good sample of the SLT99. To sort out that question would need a decent sample of each.
Another point may be that the accuracy I have measured using the linear curve on the P45+ was by any means pretty decent. Using a more film like curve may give "better" tonality although it may be less "accurate".
Best regards
Erik
I am with you on that, I always felt the standard color checker was lacking compared to the IT8. The special pigments may be hard to mix into the variety of the IT8, or at least the color brightness.
I believe your results on ACR being good. I have a small doubt on the C1 considering so many people like it. Maybe it is marketing and you have pulled the curtain back on the wizard of oz. It would be nice to know what products give a precise "neutral" color.
-
I am with you on that, I always felt the standard color checker was lacking compared to the IT8.
isn't IT8 way too glossy target to use (and wasn't that the reason why Colorchecker DC failed) ?
-
isn't IT8 way too glossy target to use (and wasn't that the reason why Colorchecker DC failed) ?
I don't know. I would think IT8 is fine in diffuse, high CRI light. I do know Silverfast says it is not designed for DSLR calibration. I settled on the idea of the IT8 as a gamut test more that a color test for DSLRs.
Imaging Resource has ISO100 raws for a huge number of cameras with a color checker in the shot (still life series). If there is a problem with a raw converter is should be easy to test.
C1 should be designed for MFDB compatibility so it is very surprising if they fail there.
-
Hi,
This is D800, Capture One using linear curve. LR 5 significantly more exact.
DPReview has an IQ180 test shot they have as reference shot. I have tested the CC with both C1 and LR5. C1 was more accurate.
From all tests I have done I would say that Capture One is pretty accurate on the Phase One shots I have tested with the linear curve, but less accurate on DSLRs. LR is easy to calibrate to each camera.
The differences are not really large. What I can see that saying that C1 and MFDBs are more accurate seems to be a blanket statement without much substance. It is very possible that they yield a more pleasant correction. Accuracy can be measured but pleasantness can not.
It would be helpful having more samples, and I will do some more testing.
Best regards
Erik
I don't know. I would think IT8 is fine in diffuse, high CRI light. I do know Silverfast says it is not designed for DSLR calibration. I settled on the idea of the IT8 as a gamut test more that a color test for DSLRs.
Imaging Resource has ISO100 raws for a huge number of cameras with a color checker in the shot (still life series). If there is a problem with a raw converter is should be easy to test.
C1 should be designed for MFDB compatibility so it is very surprising if they fail there.
-
Hi,
The SLT 99 is pretty close the Alpha 900 in SMI rating. But I will repeat the test with the A900.
Best regards
Erik
Yes, the A900 was Sony's attempt to make a photographer's camera with no frills, no BS. A large part was the choice to give up some ISO noise for color separation on the filters.
-
Hi Sareesh,
Unfortunately, no. I have two issues with skin, I have exactly two portraits shot with my P45+, one of those is showing myself, and that one is discussed here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition?start=6 .
The other issue is that as I am a landscape shooter, I cannot really say what skin looks like.
But, I have found some oddball issues with the Adobe Standard profile on the P45+, that I fixed by generating a DCP profile, see below:
(http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/Screenshots/P45+_AdobeSTD_vs_DNGProfiler_small.png)
So I would say that a profile may be reasonably accurate and still produce outliers. The P45+ with the C1 linear profile has outliers at DeltaE* = 17.2, while LR 5.3 with my home made profile has a Delta E of 7.13 for the same patch. Still, the very same patch on the SLT 99 has a Delta of 1.48.
I had downloaded a two raw files from Nikon D3X and Aptus (of some kind). In those images the D3X had a pinkish skin tone with the Adobe Standard profile, but the exposure also included a ColorChecker,
which enabled me to create a DCP profile that removed the pinkish tone. The D3X still looked flat compared to the Aptus.
The image below shows the outliers:
P45+ with capture one on the left, P45+ with LR5.3 at the center and SLT 99 on the right. The LR5.3 files use DCP profiles generated from a ColorChecker Passport.
Best regards
Erik
Thanks for sharing! From a quick peek at your photos on your site, I prefer the C1 look, but I might not know what I'm seeing so I'll read through them this week and get back to you.
-
Hi,
After doing some additional experiments here some summary.
My findings here focus Delta E* using the CIEDE 2000 formula. A possible interpretation DeltaE* may be:
- DeltaE* = 1, barely detectible
- DeltaE* = 3, visible in images
- DeltaE* = 5, objectionable deviation
I found that of my two cameras tested the Sony Alpha 99 had the lowest DeltaE* using Adobe Lightroom 5.3 with a home made DCP profile, but Adobe Standard profile was close, so I will normally use the AdobeStandard profile on that camera.
With the P45+ the issue is a bit more complex. What I found initially was that Lightroom 5.3 with both Adobe Standard profile and my home cooked DCP profile produced smaller deviations than the Studio flash profile in C1 I compared with. But, I also know that I was not satisfied with the Adobe Standard profile, the very reason I generated my "dcp" profile. Very obviously Delta E* doesn't tell whole story.
Switching to linear curve improved the statistics, with linear curve C1 produced small deviations, almost as good as the Sony.
I would presume that DeltaE* takes both Luminance, chroma and hue into account, and the film curve varies luminance intentionally. Using the linear curve I assume L stays linear/straight in some sense.
So this would indicate that linear curve in Capture One gives good rendition and my guess is that film curve just adjusts tonality.
I also made a comparison between "Portrait natural" and "Flash" in C1 and the differences were very small (way below DeltaE* = 2 on individual patches) and mostly in areas nearly skin tones. I presume that we see some of the "image professor's" hand at work. Very small subtle effects in selected areas. Delta E* mean between the two was just 0.51. I enclose the comparison between the linear profile and portrait profile.
The outdoor daylight profile and the flash profile were essentially identical, no deviation at all.
Personally, I feel I have learned a lot about the options I have at hand.
Best regards
Erik
-
Using Delta C or Delta L alone would not give any good information. C is essentially saturation, pretty meaningless without hue angle. H is hue angle (essentially). There can be large variations in hue angle for areas of low saturation without a visible difference. So Delta C or Delta H alone give little useful information.
Eric,
I think you are using ΔC too restrictively. My knowledge of color difference formulas is largely limited to those used in Imatest and explained in the supplementary documentation. ΔC* refers to differences in chroma, but other ΔCs include ΔCab, ΔC94, ΔCcmc, and ΔC00. One can omit L* with these formulas and they would be better for your purposes than ΔE. What are your thoughts?
Best regards,
Bill
-
Hi Bill,
I used "Patchtool" for this and I don't think it has a "Delta " mode which excludes L but includes color. I tested both C and H. C just says how saturated a color is and H can vary wildly. In some cases I have seen C = 0 and hue angle varying 180 degrees.
My version of Imatest doesn't handle IT 8.7 tagets, AFAIK.
I feel that a proper comparison needs to eliminate L, as it is essentially tone curve. Capture One having linear curve is really a nice feature, at least for testing by the numbers.
Best regards
Erik
Eric,
I think you are using ΔC too restrictively. My knowledge of color difference formulas is largely limited to those used in Imatest and explained in the supplementary documentation. ΔC* refers to differences in chroma, but other ΔCs include ΔCab, ΔC94, ΔCcmc, and ΔC00. One can omit L* with these formulas and they would be better for your purposes than ΔE. What are your thoughts?
Best regards,
Bill