How do you evaluate color of a new camera?

[torger]

I ask mainly as a landscape photographer. I've just got this new H4D-50 camera, only using the back in on my Linhof. Previously I've been using Leaf Aptus 75. I'm now running a series of tests to see how well the H4D-50 performs compared to my Aptus 75.

As many of you know if you use the same raw converter for two cameras and generate custom profiles for daylight illuminant (using a color checker) you will get almost indistinguishable color under that light condition. I did that experiment and got the expected result. However in landscape the light is seldom plain daylight, and here in northen Sweden we have pretty extreme light during the winter season, but also during summer with midnight sun etc.

My way to work with color in landscape is to use a daylight profile regardless of light condition, and then hand-tune white balance until it looks natural and realistic according to my memory. I may tone the picture slightly for effect, but I rarely correct individual colors.

When the illuminant changes form the profiled, color rendition will modulate largely as an effect of the sensor's CFA. That is at other light I will start to see a difference between the Aptus 75 and the H4D-50. It would take me a year to go through all seasons and my typical subjects though so it's not a very efficient way to evaluate. The little tests I've done by modulating a reference light seems to suggest that while the Aptus often wants things to be greener than they should (which is what I've felt in the field too), the Hassy will make things redder, but I don't know for sure from that little testing.

The only thing I can really conclude from my testing is that color will not be exactly the same with the Hassy, but will it be better or worse, or just different? My best guess is the latter and I'm quite comfortable with that, but I don't know, and I don't know how to find out other than start using it and hoping I'm happy with it after a year (seasons here are very different in terms of light and how subjects look, now we're buried in snow).

Any ideas?

[MichaelEzra]

Hi Anders, considering that viewing of the final image conditions largely impact its color reproduction and perception, I'd say that practically, it is only marginally important that camera's color reproduction is correct. If application is reproduction work, then it would be quite different, but for landscapes which normally would go through further (creative) color tuning anyway, I  don't think exact camera color is that important, as long as it is decent. For landscapes, factors that seem more important to me would be camera's dynamic range, noise and color depth. I am sure you will be all set with H4D-50:)!

[torger]

I'm with you Michael, I don't think it really matters much - ie color is different but you can't really say which one's better. I'm however  interested hearing if someone had some method to find out.

It's one thing if you're evaluating a camera for a fixed studio condition, easy to recreate and shoot in those conditions, but for a landscape photographer you can't really recreate the light conditions and subjects easily without a lot of field testing over a long period of time. Shooting in snow at winter with a low sun is very different from shooting in autumn colors in overcast condition.

I've heard some complain on green separation with some Kodak sensors (I think it was the H3D-39) but I'm unable to see any difference at all concerning green separation (the 50 megapixel sensor could be better of course...), on the other hand my lab test subjects (fruits and stuff) is different from moss, leaves etc, maybe there's some metamerism issue waiting to happen. I guess there's no way around shooting the actual subjects in actual conditions side by side (just as you would do with a model in a studio if you're a portrait photographer), and then it's just not feasible so it would be to relax and hope for the best. It would really be nice if I could make some color separation test in a controlled condition that actually said something. I'm writing on a review and it would be cool if one could say someting more than "it doesn't matter, all cameras do well if you take control over profiling", but I'm suspecting that it's actually the truth.

I'm not too worried though because I haven't really been displeased with any color rendition of any reasonable camera, as long as I get to control it myself. Default profiles in raw converters however is often messing things up in a direction I don't like. Phocus color looks okay except for some saturated colors which are just plain wrong, but once I apply a custom profile it looks fine, and it's the same story with Capture One and the Aptus.

[eronald]

I ask mainly as a landscape photographer.
The only thing I can really conclude from my testing is that color will not be exactly the same with the Hassy, but will it be better or worse, or just different? My best guess is the latter and I'm quite comfortable with that, but I don't know, and I don't know how to find out other than start using it and hoping I'm happy with it after a year (seasons here are very different in terms of light and how subjects look, now we're buried in snow).

Any ideas?

Torger,

 You take your favorite subject under your usual lighting, and shoot. If the color discrimination looks ok for your subject, then your camera and workflow are ok. If you have trouble discriminating colors you see as different, or the colors of the image do not match your perception, then you are in trouble. A profile made with the adopted illuminant can fix some issues, but it cannot improve the color discrimination of the camera, or fix some spikey issues that may occur in the wrong place; something always ends up appearing with every sensor. The other thing which is hard to fix is if under your usual illuminant the sensitivity in some channel is too low, and you get noise in that channel - if that happens you really need to use a filter to match the channels at capture time.

  The thing is, you need to find out whether your own color vision is ok, and to have your results checked out by someone with good color vision. The rule of thumb is that young females have considerably better color vision than older males - but there are exceptions of course

Edmund

[torger]

My color vision is pretty awesome , I've quite recently tested it. It's not awesome enough to see difference between CMOS and CCD on scaled down images and I'm not trained to evaluate skin tones (which are not interesting to me as a landscape photographer). But I can differ between tiny differences in tones in the more general case.

The sensor does have about one stop less sensitive red channel for the daylight illuminant, while the Aptus is more balanced for that. I thought that could be a problem so I've looked at that specifically. The increase of noise in reds seems in practical use to be negligible though. I'm trained to do ETTR with my Aptus, and this Hassy is also an ETTR sensor, and as long as I do my exposure right it seems like I will be fine concerning the noise issue.

My favourite subjects and lights varies throughout the very varied seasons around here, so I was thinking if one could make some lab test to make sure there's no really bad spikes or metamerisms or something that could bite me later on.

[MichaelEzra]

I suppose, that since in landscape photography subject matter is not predictable, its metamerism is likely also. The only way around that would be to create a color profile for each scenery...

I should admit that what fascinates me personally in landscape photography is freedom of color manipulation for conveying the feeling. I am yet to see a landscape photograph that would come out of the camera directly and be considered a work of art. As such, I find it appealing to consider color manipulation as an art form & skill vs accurate technical reproduction. There is also a psychological factor that colors remembered and felt are quite different from the actual colors seen. SilkyPix had an option "Memory color" that tried to address this.

[ErikKaffehr]

Hi Anders,

I would say that getting the WB right is most crucial, and not at all easy in mixed light…

Personally, I have tried the usual suspects ColorChecker Passport and Adobe DNG Editor. Added to that QPCard and it's own software. I have been spending some time on this recently as Tim Parkin of OnLandscape suggested I write a small article about the issue.

The way I test, is essentially shooting an IT-8 target under controlled conditions and comparing with the reference data that came with that card. In real world I have a set of profiles generated with each.

I would guess that dual illuminant profiles work probably well for continuous spectrum illumination, but spiky spectra need a special profile.

Evaluation of the images I do with the "scanin" program from Argyll CMS, which I use to generate a CGATS file that I can read into Patchtool  (from Babelcolor). Patchtool can make nice comparisons, both visualising colour differences and calculating Delta E* values and statistics.

The reason I use the IT-8 target instead of ColorChecker is that it has many more patches and covers a large part of Adobe RGB.

If the ColorChecker base calibration route works or not may depend on colour preferenses. I am happy with the colour rendition of my Sonys, and I can come pretty close with the profiles I have tested.

I attach four images, the first two show the colour errors in six different cases:

Sony Alpha SLT Adobe Standrad	P45+ Adobe Standard	P45+ AdobeDNG Prfile Editor
P45+ QPCard	P45+ ColorChecker PP	P45+ Capture One

The third images show the deviations from my Sony Alpha 99 SLT using the ColorChecker Passport Profile while the fourth one are the deviations for Capture One. So ColorChecker Passport gives a rendition on the P45+ which is close to my Sony Alpha 99SLT, while Capture One produces different colours.

Capture One pushes some of the deep blues outside Adobe RGB, so that patch gets a DeltaE* of almost 17, but it looks perfectly OK on screen because of gamut clipping. Took me a while to figure out.


Best regards
Erik

Ps. The first draft of the article I am working on is here: http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/OnColor.pdf

I ask mainly as a landscape photographer. I've just got this new H4D-50 camera, only using the back in on my Linhof. Previously I've been using Leaf Aptus 75. I'm now running a series of tests to see how well the H4D-50 performs compared to my Aptus 75.

As many of you know if you use the same raw converter for two cameras and generate custom profiles for daylight illuminant (using a color checker) you will get almost indistinguishable color under that light condition. I did that experiment and got the expected result. However in landscape the light is seldom plain daylight, and here in northen Sweden we have pretty extreme light during the winter season, but also during summer with midnight sun etc.

My way to work with color in landscape is to use a daylight profile regardless of light condition, and then hand-tune white balance until it looks natural and realistic according to my memory. I may tone the picture slightly for effect, but I rarely correct individual colors.

When the illuminant changes form the profiled, color rendition will modulate largely as an effect of the sensor's CFA. That is at other light I will start to see a difference between the Aptus 75 and the H4D-50. It would take me a year to go through all seasons and my typical subjects though so it's not a very efficient way to evaluate. The little tests I've done by modulating a reference light seems to suggest that while the Aptus often wants things to be greener than they should (which is what I've felt in the field too), the Hassy will make things redder, but I don't know for sure from that little testing.

The only thing I can really conclude from my testing is that color will not be exactly the same with the Hassy, but will it be better or worse, or just different? My best guess is the latter and I'm quite comfortable with that, but I don't know, and I don't know how to find out other than start using it and hoping I'm happy with it after a year (seasons here are very different in terms of light and how subjects look, now we're buried in snow).

Any ideas?

[DanielStone]

Torger,
Why not just shoot a frame holding a grey card in front of the lens? That way, you have a reference of a neutral point, technically.
Ya, it cuts your card space in half, but hey, cards are cheaper than gas or plane flights these days. And the post work becomes simple and quick.

just my 2¢

[eronald]

My color vision is pretty awesome , I've quite recently tested it. It's not awesome enough to see difference between CMOS and CCD on scaled down images and I'm not trained to evaluate skin tones (which are not interesting to me as a landscape photographer). But I can differ between tiny differences in tones in the more general case.

The sensor does have about one stop less sensitive red channel for the daylight illuminant, while the Aptus is more balanced for that. I thought that could be a problem so I've looked at that specifically. The increase of noise in reds seems in practical use to be negligible though. I'm trained to do ETTR with my Aptus, and this Hassy is also an ETTR sensor, and as long as I do my exposure right it seems like I will be fine concerning the noise issue.

My favourite subjects and lights varies throughout the very varied seasons around here, so I was thinking if one could make some lab test to make sure there's no really bad spikes or metamerisms or something that could bite me later on.

Anders - I think the camera you have is well known here - just ask people what its strengths and weaknesses are!
It's always time to roll out the experimental equipment later ...

Edmund

[ErikKaffehr]

Hi Anders,

I shot an image yesterday, with my article on my mind. This was shot around mid of day, with sun still very low (although I guess much higher than what you use to have these days). It is rendered with different profiles in LR5 and also in Capture One.

This has been grey balanced from a colour checker card.

The QPCard profile is probably outside it's best range.

http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/NykopingCastle2/

What I would say in general, after checking out the conversion in ColorThink2 I have the impression that both ColourChecker Passport and Capture One push saturation on blues and greens, while Adobe's profiles are less saturated.

Best regards
Erik

[torger]

Thanks for the insightful replies.

I realize that the only reasonable way to present a test of the color is to shoot in various real conditions, which for landscape means long-term testing. Maybe I'll do at least one test for the review, I guess I can find something outside that's not buried in white snow... now I just need to get that replacement battery.

On the philosophy of color in landscape photography I do like realism as a starting point. Grey card + white balance won't work for realism though, as light temperatures are so extreme that the eye/brain system does not fully adapt. Sure in overcast conditions it works, but if you have a landscape with part air light and part low sun, the eye will see a blue/yellow mix. Dual hand-tuned white balances is often best in those cases to get the most realistic result, but I'm no fundamentalist and do with one nearly always. Only in some tricky conditions in snow I've had to use two white balances (one for the sunlit parts, another for the skylit parts) to get something that doesn't look wrong.

I use the word realism instead of accuracy as there are no known methods to actually match color with the eye's experience in extreme color temperatures. Accuracy in reproduction terms only work in daylight-like conditions. To make something look real absolute accuracy is not that important, a reasonable contrast and saturation is more important. If there actually was a method I would probably use it.

I think that the typical quote "I capture what I feel rather than what I see" has become a cliché, an excuse to do any type of manipulations to "improve" on reality. I'm quite good at seeing what type of post-processing that's been done, which I guess most photographers are, and frankly I think it's way too common in landscape to simply just put in any post-processing there to make the image "pop", to make that wow-factor that makes the image popular on social networks. I don't mind people doing that, but to me that's more about marketing than art. I've been there but felt that I was trying to satisfy an audience rather than making art that came from within. Now I'm more into starting with realism and then doing subtle grading (with cinema-like techniques) to enhance the feel in a direction that feels right. I feel that some images is best served without grading though, so I don't agree that a realistic image can't be art. However, I do agree that absolute accuracy is not important, although an interesting challenge for technology and psychovisual research (I don't know of any substantial psychovisual resarch that's been made to investigate how the eye/brain responds in extreme light temperatures).

[torger]

Regarding profiles, today with the Aptus I use a basic single-illuminant color checker passport-generated profile from real 12 o'clock clear summer day sunlight. I think it provides better starting color than Leaf's default profiles, excpet possibly for their oddly named "ProphotoRGB" (which has nothing to do with the color space with the same name) but is their "neutral" profile. I'm sure the default profiles are great for studio portraits though, but I see no reason to use them as basis for my landscape work. I've come back now and then to see how they render a particular image, but I've always returned to my custom profile.

The passport generated profile seems to have problems with very saturated colors. It has not been a big issue as those colors are rare in my photography, but I still think about making a more advanced profile with some other method. I don't think a very low deltaE is most important. It's more important that the full range is "within reason" than extremely low deltaE in some range and large in another. I also suspect that it's important that the profile does not become too nonlinear and spikey which might cause gradient problems.

It seems to me that there are not much tools on the market for making profiles though, it seems like there's an opportunity to make new software products in that area. The lack of good understandable tools I think is one reason why manufacturer's own profiles has got it's "black magic" status, photographers in general don't really dare to make their own all-around profiles because they don't think they can succeed as good as the manufacturers. I'm not so sure on that. Sure, you won't have a monochrometer as a layman, but I find it hard to believe that it's actually necessary to make good profiles, especially when it comes to making a profile which is about pleasing color rather than reproduction quality accuracy. What I am sure though is that Phase One, Leaf, Hasselblad and Adobe etc have their own custom-made software tools to design their profiles, and I guess the software that's on the open market today is not as good when it comes to subjective profile design.

[torger]

Narrowing down what I would like to know about a camera's color performance for landscape applications I guess that would be metamerism on typcial subjects in nature (if different colors are registered with the same raw RGB values no profile can fix that), and some sort of view of how colors will modulate for various light temperatures available in nature. Possibly one camera will modulate colors in a more natural and realistic way than another.

A good camera would produce pleasing results with a daylight profile and just white balance adjustments, while another camera might require more custom pulling of colors. I don't really know if there are any significant differences there, but it would be interesting to know and have a repeatable method to find out.

[torger]

I have a decent photospectrometer so I can measure illuminants, but I have no means simulating one. I could do a number of interesting experiments if I had that, but I don't know if there is any possibility to do that without spending like 100k USD...

[MichaelEzra]

Hi Anders, as I was thinking about your question, I share the pain of not being able to find any single camera profile that successfully worked on the broad range of tones and colors when photographing nature as it in fact is the most extreme and subtle subject matter. There is clearly a gap here and as you said may be an opportunity. It would be invaluable to have a silver bullet.

I expect, however, that there are some fundamental limitations which are characteristic to our vision. Not only that neither of us can explain how red is the red that we see but also that color perception depends on the neighboring colors in the image. As different viewers would pay varying subjective attention to image elements, the color perception of the scene would vary also. The notion of realism then would be a very fine line. It would be the art of (yet to be perfected) science or art of visual arts to achieve it. Quite non trivial but possible. As you said, it is rare to find subtle and non-exaggerated landscape works, but when they do look hauntingly real they leave a memorable impression.

[B-Ark]

Torger,
Why not just shoot a frame holding a grey card in front of the lens? That way, you have a reference of a neutral point, technically.
Ya, it cuts your card space in half, but hey, cards are cheaper than gas or plane flights these days. And the post work becomes simple and quick.

just my 2¢

A good idea, however ...
If one is photographing a landscape that's a mile away, then the light falling on the landscape is not necessarily the same as the light falling on the photographer.

[ErikKaffehr]

Hi Anders,

Just some toughts…

1) Something that I have learned at Hans Kruse's workshop that colour temperature can be used locally. In some cases it may be essential. For instance in a sunset subject the shadows may be illuminated by pure skylight and may be well above 9000 K while the skyline may be at say 2800 K.

You wouldn't compensate for any of that, it a part of the subject.

2) It can be the reflected spectrum can be quite tricky for some subjects. Skin tones for instance have a very steep rise 430 nm and 580 nm. Tim Parkins says chlorophyll has a tricky spectrum, I guess quite different from skins. So, it could be that some sensors may be optimised for skin tones and others for greens in nature which are mainly chlorophyll. Just a reflection…

3) Kodak and Dalsa publish sensivity curves for their sensors.

I don't think you need to pay 100 kUSD for a monochromator. Theres is a device from ImageEngineering that costs just below 1k€

http://image-engineering-shop.de/shop/article_IE-camSPECS%2520express/camSPECS-express.html?sessid=shh1SLs40K5itL1P352YtVsI0x0H5s6XuEZRymTobDrNVoHvIbp0HuTHmxzEWW2A&shop_param=cid%3D565%26aid%3DIE-camSPECS%2520express%26

But I guess that is more than what you or I would be willing to spend.

Best regards
Erik

Narrowing down what I would like to know about a camera's color performance for landscape applications I guess that would be metamerism on typcial subjects in nature (if different colors are registered with the same raw RGB values no profile can fix that), and some sort of view of how colors will modulate for various light temperatures available in nature. Possibly one camera will modulate colors in a more natural and realistic way than another.

A good camera would produce pleasing results with a daylight profile and just white balance adjustments, while another camera might require more custom pulling of colors. I don't really know if there are any significant differences there, but it would be interesting to know and have a repeatable method to find out.

[MichaelEzra]

A good idea, however ...
If one is photographing a landscape that's a mile away, then the light falling on the landscape is not necessarily the same as the light falling on the photographer.

That would be a perfect excuse to use a drone!
A global uniform distribution of drones or on-demand with the white card or a color checker for a small additional fee:)
In fact, photo drones should have gray cards and color checkers built in:)

[ErikKaffehr]

Hi,

I like your ideas about the drones… Until we see it workingI guess we need to estimate colour balance, use grey card or depend on the WB in the camera.

I guess some of the problem is that we have a scene and a perception of that scene. The task is to convert our raw data to an image that corresponds to our perception.

Here is an image I have been struggling with:


"Ken R" came up with this interpretation:


"leallen" with this one:



Chris Livsey had another view:


Quite some learning experience.

Best regards
Erik

That would be a perfect excuse to use a drone!
A global uniform distribution of drones or on-demand with the white card or a color checker for a small additional fee:)
In fact, photo drones should have gray cards and color checkers built in:)

[Jim Kasson]

The sensor does have about one stop less sensitive red channel for the daylight illuminant, while the Aptus is more balanced for that.

That's quite a difference. In fact, it's a large enough difference to make me suspect that the camera with the higher red channel is responding more strongly to IR. It's pretty easy to check; slap an IR-cut filter on each camera and see what happens.

If it's not IR sensitivity, and you want to make camera A respond to spectral input the way camera B does, my first step would be to experiment with filtration. CC filters are the most readily available. Note that you can get effects with filtration that are unobtainable after capture, no matter how much profiling you do. Once two different spectra resolve metamerically to the same RGB triplet, no profile can separate them again.

Filtration also has the potential to mitigate the illuminant capture metameric errors you mentioned in your first post.

Worth a try, anyway.

Jim