I recently added support for digital camera raw files to my photo editing software I've been working on, using LCMS. I too am using DCRAW to get to the actual raw data. Here's what I've concluded up to this point, concerning profiles:
 
It all depends on whether you scale the raw data (after Bayer conversion) to some reasonable range before profiling, and it depends on what software you use to build the profiles. And it's imperative you use 16 bit color channels.
 
I purchased an IT8 target specially built for digital camera profiling from the folks at Picture Window Pro, as part of their Profile Mechanic package. However, I was not able to build a decent profile for my Canon 20D camera with that package unless I scaled the image first, and even then the resulting profile didn't work very well. It had a lot of posterization, and other such problems. And if I didn't scale the data first, the resulting profile was truly atrocious. I was however, able to profile both my flatbed Canon scanner, and Nikon 8000 film scanner with great success. The profiles were very good. But the digital camera profiles, sucked, to put it lightly.
 
So as a goof, I tried using the free AIM profiler (from the Timo guy) written in Excel. I wasn't expecting it to work, but actually it worked quite well! While I was able to build a profile either with or without pre-scaling the data, I got a better profile when I scaled the data such that the maximum value encountered in any channel was set to 0.95, on a scale of 0 to 1. This gave plenty of highlight headroom to work with when doing color balancing later on, but also gave the profiling software reasonable data to work with. It's imperative you work with at least 16 bit channels at this point, or you'll have no hope at all of getting anything decent. Also, I find it best not to touch the shadow point at all at this stage.
 
With the resulting device profile, you can at least get a reasonable looking image in your favorite photo editor. You'll still probably have to tweak the white points, color balance, etc, but the image will be in the ballpark. While I'm still experimenting, I've found that it's best to set the color balance and white point while in some sort of linear space. From what I understand, most digital cameras are fairly linear in their response to light, and by creating a device profile with AIM for example, you basically are compensating for any deviations along the straight line. I believe the AIM profile does not apply a "gamma" at all to the data, but I may be wrong about that. (In other words, any gamma compensation happens in the conversion from the device profile to the monitor profile.)
 
I seem to be getter better results doing the color balance in the device profile space, but that's not conclusive. (Can anybody shed any light on that?) Then I convert to a linear space based on ProPhotoRGB primaries or some other wide gamut space, but I don't use the 1.8 gamma of ProPhotoRGB, but rather, stay in a linear space. From there, I set the white point and basic midtone brightness. HOWEVER, I do NOT set the shadow point in linear space. It's way too senstive, as all the shadow values are jammed up next to each other. Convert to a profile that has a gamma of 2.2, or thereabouts, and THEN set the shadow point. Note that you should only convert to 8 bit AFTER changing the gamma. I've also discovered that it's sometimes better to set the midtone brightness after changing the gamma. It seems to pack more punch in gamma 2.2 space than in linear space. I'm not sure about that yet, though.
 
I've used this workflow successfully with a Canon 20D and also a Canon G2, but I'm still experimenting and there's a lot to learn.
 
As others have mentioned, the problem with using profiles that aren't built specifically for your DCRAW processed data, is that you have no way of knowing what assumptions are being made about scaling, gamma, etc, when those profiles were built. I've found that using generic profiles I've gotten of the web, or from, for example, Capture One, result in dull lifeless and sometimes incorrect color. There's nothing wrong with these profiles per se, but they were written with different assumptions in mind than what I use in my  DCRAW converted data. I haven't been able to determine what those assumptions are.
 
Bryan
 

Reply via email to