Re: Camera Profiling
Re: Camera Profiling
- Subject: Re: Camera Profiling
- From: Ben Goren <email@hidden>
- Date: Thu, 29 Jul 2010 08:00:09 -0700
On 2010 Jul 28, at 9:05 PM, Lindsay Merritt wrote:
> I decided to build a custom profile for the 7D.
Lindsay, I feel your pain. And so does anybody else who's tried to get colorimetrically-correct results from a digital camera.
If you don't already have a spectrophotometer, start budgeting now for an i1 Pro (or at least the Munki). It will become an indispensable tool. More on that later.
The best workflow I've come across so far is to use well-controlled lighting; to use a manually-linearized DNG profile with the RAW conversion; and to use ArgyllCMS to build an ICC profile as well as to convert the profile to the working space.
In just a little bit more detail:
I recently got a Sekonic meter. Out of the box it was pretty good, but the incident and spot readings didn't quite match. The i1 has an incident light measurement mode, and Argyll will report the illuminance levels in lux when you use it that way. To convert from lux to EV(ISO100):
EV = log2(lux / 2.5)
The Sekonic was a couple tenths of a stop off from the i1. Once I adjusted the Sekonic to match the i1, the ambient and incident readings on the i1 were a perfect match, and both matched the in-camera meter.
Next, I photographed a ColorChecker with just strobe modeling lights, set up in a copy-stand configuration. I adjusted the lights until the meter registered an exact stop setting, to the 1/10th, across the entire field. I set the camera to match the meter reading.
To get the perfect white balance, don't use the eyedropper. Instead, crop the picture so just your white balance target is visible. Then, use the exposure adjustment until the histogram is somewhere near the middle (it doesn't need to be precise). Lastly, adjust the two white balance sliders until the histogram is a single white peak. In ACR, you'll probably need to type in the actual color temperature, as the 50K intervals the slider and arrow keys work in aren't very fine-grained. (Side note: DPP always shows the histogram of the entire picture, not just the cropped portion. I emailed Chuck Westfall asking him to ask The Powers That Be if they'd fix it, and he said he'd pass on the request.)
A bit of a diversion here: for these purposes, Tyvek is the absolute best white balance target you'll find, short of Spectralon (a suitably-sized piece of which will set you back more than the camera body). Tyvek has a near-perfect flat 99% reflectivity across the entire spectrum. It's not quite as good as PTFE, but the difference isn't something you can measure with photographic equipment. (Spectralon is an almost-perfect non-specular reflector as well as being 99.9% reflective across the visible spectrum, whereas Tyvek's surface is somewhere between a matt and semi-gloss photographic paper.) The best thing about Tyvek? It's dirt cheap. You can buy Tyvek mailers at your local office supply store, and you probably already have a Tyvek CD envelope laying around somewhere. If it's a natural white plasticy paper with lots of fibrous detail, it's probably Tyvek. If a Sharpie pen spreads like mad along those fibers, it's almost certainly Tyvek. And if a plot from your spectrophotometer shows a flat line a hair's breadth below 100% reflectivity, then, congratulations, it really is Tyvek.
(Canon sells Tyvek in roll form for banner printing. But, before you get too excited...the printable side is coated with the same FWA-rich stuff they use for their generic matt paper. The back side, however, is uncoated and shows no sign of fluorescence or other adulteration. So, if you do banner prints, get a roll, use it for its intended purpose, but also cut yourself a large blank sheet that you spray mount, back-side-up, to some foamcore for your perfect white balance target.)
Next, remove the crop, return the exposure to neutral, and set all the other controls to their neutral (*not* default) position. In particular, brightness and contrast need to be at 0, the tone curve needs to be flat, and so on. Save these as your new defaults.
At this point, in ACR, if you use the Camera Faithful DNG profile, you've got a not-miserable starting point for profiling, but we've still got a long ways to go.
From here, I opened the profile in the DNG profile editor. Though I used the chart feature to adjust the (2600K) color table, what I was really interested in was the tone curve.
You want to start with a linear curve as the base curve, and then adjust it until the six neutral patches have the same Lab values as the ColorChecker. How to do this...is far from obvious. You'll need to know the Lab values from your ColorChecker, of course. You can find averaged values online, or you can measure your own ColorChecker with your spectrophotometer. (I did mention it'd be indispensable, no?) Unfortunately, the DNG editor doesn't give you a readout of picture values, and this isn't something you can eyeball. But...Mac OS X ships with a ``Digital Color Meter'' that will read out in Lab values. It takes the RGB levels sent to the video card and runs them through the monitor profile.
Be sure you have the DNG editor set to apply CameraRaw adjustments, and then create a curve that gets the Digital Color Meter readout of the patches to match. Set aside at least an entire afternoon to do this...there's so much trial-and-error and frustration involved that I can't offer any specifics other than that it *can* be done, and that the simpler the curve the better. The end result should be very smooth, and all the Lab values should match to within a couple L* units. But be patient and persistent.
Next, you want to calibrate your flash meter. Adjust the flash heads until the meter reads an exact stop. Take the shot, and see if the N/ 5 patch on the ColorChecker matches what it should (using, of course, your brand-new DNG profile and all settings at neutral). If not, change the calibration on your flash meter appropriately, re-measure and re-adjust the flash heads, lather, rinse, and repeat until the flash meter is properly calibrated.
Then, use the shot of the ColorChecker lit with flash to create the 6500K color tables in the DNG profile you created earlier.
(I found the auto-adjustment of the color tables to only marginally improve results, but it did improve them. My ColorChecker Passport arrived yesterday, and I hope to get a chance to put it and its software through its paces sometime today or tomorrow.)
Now, we *finally* get to the point of creating an ICC profile. Shoot the best profiling chart you've got. (I'm actually in the process of making my own, with dozens of paints -- artist's acrylics and paint store both -- applied through a silkscreen.) Be sure to use the exact same lighting conditions as you intend to use for real. Apply your shiny new DNG profile, save it as a ProPhoto RGB TIFF, and run it through Argyll to create a profile. (That last step will be non-trivial, but it's well documented and the support on the mailing list is fantastic.) (Oh -- and don't worry about the fact that it's already tagged with ProPhoto. It's a lengthy explanation, but the short version is that the profile it's tagged with is entirely arbitrary, and that's just the best to avoid clipping in the RAW converter. Argyll will take care of changing ``what it is'' to ``what it should be.'')
Finally(!), shoot whatever it is you were planning on photographing in the first place, save it using the same workflow as you used to create the image of the profiling chart, and use Argyll to convert the image to your favorite working space. (You could simply assign the camera profile in PhotoShop and let it convert it to the working space, but Argyll does a better job. Especially if you do gamut mapping and a device link profile, though that's somewhat slower.)
Obviously, once you've done all the preliminary work, the actual production work can be automated. If your lighting never (significantly) changes, you can reuse the same ICC profile for forever. Batch process the RAW files to TIFFs, and Argyll was designed from the start as something to be scripted.
Is this an insane process? Yes. But it gives the best results of any I've found so far. Whether or not it's worth it is another matter...but, based on your email, I'm pretty sure you'll decide it's at least worth a try.
An almost-final note: the state of the art in fine art reproduction is way beyond ICC profiles. Roy Burns at art-si.org is doing mind-blowing stuff, including taking multiple shots through different filters and using the result to create a spectrally-represented (rather than RGB-represented) file. And he's not the only one working on this kind of thing....
A really-final note: I've found this linearized DNG profile to give me the best starting point for ``artistic'' photos as well as copy work. Generally, I only have to make very subtle adjustments to the black point or contrast to created the desired ``punch,'' and everything else after that is dodge-and-burn. If I could nail the lighting in the real world (obviously not possible with available light in the field), I wouldn't even have to do that much. The built-in profiles / curves / etc. have more ``punch'' out of the box, but they also clobber details and destroy color fidelity, and you can't get it back. It's much easier to start with something that's as close to the actual scene as possible and then adjust it to your specific tastes than it is to try to start from somebody else's interpretation and to re-re-re-interpret it to your liking. At least that way you get to be the one to decide which details to clobber and which colors to alter.
Cheers,
b& _______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden