James ...

As an experiment only, it is possible to convert one of your "skin tone portrait" raw images to DNG, and then to open it with the DNG Profile Editor (DPE). After you've read the tutorials regarding editor's use, you can play with the editor a bit to try and make colors better ... and then to describe here what you've done and Eric Chan may be able to say more.

Altho I do have to admit that something may be wrong with your color management if the skin tones look worse for profile 'portrait' than they do for profile 'landscape'(?) IE, it could be your monitor profile that is causing skin tones to "appear" off.

My camera isn't Canon, but I was very impressed with the color accuracy of the standard beta profiles.

The Xrite colour checker is also not supported (unless it's a direct knock
off of the gretag one)




Gretag Macbeth does not exist any more. They were acquired and absorbed by Xrite. The Xrite color checker is the Gretag Macbeth color checker.

See, for instance:

In general if you're seeing a color cast (e.g., too green) the issue is white balance, not the profile. A simple tweak of the WB should fix it.

I'm not exactly sure what you're looking for, though. In general you don't want to get the exact colors that the camera saw because then you're talking about scene-referred color and that's downright ugly. You generally want something that's been tone mapped in a way that makes the image look pleasing, and the way you do this could vary image by image.

A single profile, no matter how "accurate" it is, whatever that means, is highly unlikely to produce desired results across all images because the colors you're looking for tend to be image-specific and content-specific.

Ramon; My apologies, I wasn't aware they'd been purchased. That's good to know, the Xrite cards seem cheaper than the Gretags were.

Eric; I don't believe my white balance is incorrect necessarily... different profiles GIVE the image a very slight colour cast because they interpret colours in a certain way. For example the Landscape profile intentionally saturates skies because everyone wants their landscapes to have bright blue skies.

What I am looking for is exactly as you say, "exact colours that the camera saw" which I wasn't aware was called "scene-referred colour".

I'm interested in my base starting point being 100% exactly what the camera saw, atleast in so far as a 40D can reproduce those colours. Once I've got that starting point where the colours are as accurate as possible (assuming the white balance is right) then if I find the result undesirable then I can tweak the saturation or tweak the HSL sliders etc but I don't want to be tweaking the profile.

Correct me if I'm wrong, but isn't that the whole point of the auto-calibration option in the Profile Editor?

Yes, PE should give you a relatively accurate and consistent starting point, within technical limits depending on your lighting conditions.

BTW, James, if you're really interested in using the auto-calibration feature of PE, you can get a miniature version of the ColorChecker. (I think X-Rite has a small version for less $$$. It'll work just as well as the larger, more expensive versions.)

Hi Eric,

Yes, I'm looking into these at the moment (thanks to Ramon for clarifying that they were compatible with PE).

I bought the Perfect Pixs Natural Colour Card (30 patches) because it had more skin toned colours for calibration, but it's no use to me now.

Cheers,
James

I've been fiddling for the first time with the 24 patch Color Checker trying to get my Pentax K100D PEF's to reproduce exactly what I saw (not what the camera saw) when I trip the shutter. I've been studying and researching this through trial and error for some time now and I've come to the conclusion that one profile can't fit all scenes. I have to tweak each one for subtle variances.

I've compared two converters ACR 3.7 (also 4.0-4.5) using Tindeman and For's ACR Calibration scripts and Raw Developer (Mac only) using its matrix based ICC profile generator in combination with its Lab curve feature and I run into color errors due to different spectral reflectance properties of real scene objects and lighting situations.

I'm very picky about getting exactly what I see that the differences wouldn't be noticed by most folks. For instance shooting up close light gray brick wall with yellowish beige mortar lit by 10AM 45 degree direct sunlight renders all one warmish hue in both converters where the brick should be cooler and the mortar should be warmer. Color temp adjust just makes everything cooler or warmer. This is where the DNG Profile Editor would come in to fix this.

But also I found changing the angle of the camera to the subject and underexposing by one stop fixed this instead of tweaking software. I mean there's no way of getting any consistency unless you're shooting under controlled studio lights and I'm not even sure about this since I haven't tried it.

The thing I've discovered with my Pentax about getting exact numbers with the Color Checker is the exposure needs to be set so that the white square is already close to being between 230-245 RGB when first opened using the default raw converter contrast curve. I've noticed that after both scripts are run that the scene the CCchart was shot in has now lost quite a bit of contrast. The numbers are dead on to the published data and so are the look of the colors but the rest of image doesn't look like my eye saw the scene in regards to contrast. I believe this is due to the linear properties of the instrument used to measure the chart to arrive at the published numbers. When I correct the contrast, all the numbers are off looking noticeably saturated but not by much.

What these scripts and the PE wizard do is make sure your hue's are as they should be separate from density adjusts which affects saturation. I've also noticed that lightening saturated colors by applying a slight curve tweak will noticeably reduce saturation without noticeable change to density and contrast. PE can be used to fix this as well.

Hi Tim,

I think I understand what you mean about the saturation not directly affecting the hues and certainly I don't like what some of the profiles do to saturation for this reason.

When you say that one profile can't meet all needs, I assume for the most part this comes down to limitations in our cameras and their inability to replicate the colours we see? Do you think though, that it's possible to get a profile that's "most accurate" and then just tweak HSL sliders on a per photo(s) basis rather than end up with numerous different profiles? Or maybe a tweaking of the contrast curve?

My end goal is to apply a single profile on import to all my images... after that I don't mind changing specific things, but I want to import all my images with the best possible base settings.

Cheers,
James

PS, looks like I'll be buying another colour checker and doing much the same tests as you have.