p.2 #1 · Anyone here profile their cameras? Is it worth it?
BTW, both the samples above are actual testshots sampled with Heidelberg Prinect software and original colour patches were measured with an industrial quality spectrophotometer, and translated into continuous tone wedges. The custom profile is not even very "sharp" as I prioritized usability over accuray in this one to make it more versatile. I could without any work at all make it a lot more exact.
p.2 #3 · Anyone here profile their cameras? Is it worth it?
What the comparison doesn't show you, and what most people don't (or can't) see is that this is not "solvable" by changing white-balance. In fact WB will make the problem WORSE in 99 cases of 100.
Skintones are situated somewhere to the right of the middle of the wedges above (but they are weaker - containing more pure white). We usually WB for the skintones if there are people present in the photograph.
In the uncalibrated wedge you can see that this means that you have to increase red and decrease green (same as lowering WB-temperature and adding a slight magenta "tint" on the second slider in most raw-converters). Now we have corrected for the skintones, they seem "real".
But what have we done to THE REST of the picture? We've increased red power over green to get the skintones "right". As an effect of this, red-orange will now be mostly red. Medium red will be strong bright red. Greens will be more bluish green.
Any Canon users recognize this scenario? :-)
p.2 #4 · Anyone here profile their cameras? Is it worth it?
For a little disambiguation: People are talking about two different profile types here: ACR camera profiles and ICC camera profiles.
ICC camera profiles are specific to the lighting, lenses, ISO, dynamic range of the camera. They're a capture state of the camera and scene. For this reason they're generally only useful that have an extremely controlled lighting and camera setup. If you change the light, you're changing the response of the color that the camera sees. This is fundamental. These profiles are created by shooting a ColorChecker (DC) and then running that photo through software such as GretagMacbeth ProfileMaker. An ICC profile contains tables of MANY color combinations. ACR profiles work quite differently.
Adobe Camera Raw profiles are not ICC profiles, though the process to create them seems very similar these days. In response to one poster, the reason for creating an ACR profile under two different lighting conditions (tungsten and daylight) is *not* because the sensor behaves differently under these two conditions, but because of the way ACR engineers chose to solve the large difference in color response over these two light sources. ACR interpolates between tungsten and daylight. This is great -- otherwise you'd be profiling for every single color temperature and every light source. Obviously not what you want.
My understanding is that by generating the ACR profile under these two lighting conditions, you're giving the ACR engine the information it was engineered to use. Unlike an ICC profile, you're not correcting for a multitude of color possibilities. You're telling the ACR engine what the response is to the red, green, blue primaries under two very different lighting conditions so that ACR can correctly correlate between the two.
p.2 #5 · Anyone here profile their cameras? Is it worth it?
South: Yes they are (almost) totally different creatures, but you got it the wrong way. ICC is the more "handicapped" profile by a VERY long way. ACR (.dcp) profiles actually contain a lot MORE information (and more fine-grained too!) about what and where to correct in the colours than an .icc profile.
The dcp contains (optionally two of each, where you interpolate between the "low WB temperature" and the "high WB temperature" versions to get the values to actually use for shot):
A lighting matrix, a calibration matrix, a translation matrix, a "pre-exposure" colour-correction LUT of unlimited resolution, a "post-exposure" LUT of unlimited resolution (and a tone curve).
An .icc only contains three of these. Light, calibration and pre-exposure LUT (and a tone-curve).
And NO, the sensor actually behaves very differently if you compare 2800K light to 6500K light. The colours in between the primaries shift quite a lot, and the values needed to calibrate them "right" changes quite a lot. In 2800K light, orange needs to be shifted THAT way, in 6500K light orange has to be shifted the OTHER way if you want the correction to give you the correct colour - to use an easy example.
The ACR profiles are a lot MORE than the icc profiles. Icc profiles are meant for ONE media, under ONE restricted set of circumstances.
p.2 #7 · Anyone here profile their cameras? Is it worth it?
Dear "theSwede",
Thanks for your post. I had never seen that Prophoto calibrated wedge example before. Would you be willing to share the 5D2 profile you used in the sample? I'd like to compare it to my own DNG Profile editor results for the 5D2. The actual profile would be nice, but the actual recipie would be more interesting.
p.2 #8 · Anyone here profile their cameras? Is it worth it?
We agree here. I didn't imply that ICC was better. The ACR profile method makes more sense for many reasons. With ICC I pointed out that "they're generally only useful that have an extremely controlled lighting and camera setup" and I think that's a deal-breaker when it comes to anything outside of studio work.
Regarding the sensor changing: How does the sensor behave differently based on the type of light? I would think that it's the light that behaves differently, not the sensor. The sensor is rather 'dumb' to this and simply gathers what is thrown at it. Changing the sensor would be akin to ISO where you're physically changing the sensitivity. Am I off-base here?
p.2 #9 · Anyone here profile their cameras? Is it worth it?
South:
No, you're not "off-base"... :-) But there's quite a lot to "add up" even before any light hits the sensor surface! Maybe I should have written "sensor measurement result" in stead.
Sensors measure RGB values that are "post-projected". This means the raw data you end up with on your memory card is a result of multiplying several arrays of wavelength information curves together - illuminant spectrum, spectral reflectance of the target or "target colour", and the colour filters' transmittance. You basically have three arrays of values multiplied/integrated together into the raw values.
Change the illuminant spectrum and the raw values internal [R, G, B] distribution changes. WB by "light-temperature" implies that the light has a maximum power at "this" wavelength, and less at others - both above and below. The filters are constant. The target is constant.
Now what happens if we take nice, even, target reflectance (pure green?) and light it with very "warm" coloured light as 2800K? The green in itself has a response curve that looks like a symmetrical rounded peak centered at say 540nm, and the light has an intensity curve that that slopes down. Everything on the "warmer" colour side of the green peak will be amplified, and everything on the "colder" colour side of the peak will be attenuated. The peak is no longer symmetrical - it has a shallow slope at one side, and a steep slope at the other side.
Consider what happens to an orange target (mid between green and red) when the slopes behave like this. Raw "equal green and red" means one wavelength, one colour, under one lighting condition - but asymmetrical slopes will move the colour that gives "equal red and green" further towards either red or green. The profile has to correct for this. 50Red,50Green, 0Blue means one target colour in front of the camera in one lighting condition, and another with different light (even after WB).
This is what the temperature dependent LUT calibration is about. You not only have to account for different total strength of the R,G,B channels, but also their effect on the colour in between the primaries. And this only works when the light is "well behaved", like a light bulb that glows at a steady 2800K, or any other light with a black-body radiance spectrum like halogen and to a certain level, speedlights. Fluorescents are VERY asymmetrical, and "spiky" in their spectral emittance, so they're usually the hardest to profile as [light spectrum] x [target reflectance] can vary WILDY in very short distances - 550nm might be 3x stronger than 560nm...
So the sensor/system behind is indeed very "dumb", but it has to make some very intelligent decisions to get you the right colour.
Most dcp profiles that still incorporate the original Adobe temperature dependent initial correction tend to be usable over quite a large range of WB values, when I calibrate for D55 light that profile usually shows very small deviations in any lighting between 4500K and 7000K.
redcrown:
I don't have the profiles here, but I'll try to remember posting a link later tonight.
