Re: Feedback on success, creating a camera profile
Re: Feedback on success, creating a camera profile
- Subject: Re: Feedback on success, creating a camera profile
- From: Andrew Rodney <email@hidden>
- Date: Sat, 04 May 2013 17:23:19 -0600
On May 4, 2013, at 10:31 AM, Ben Goren <email@hidden> wrote:
> Indeed, right at the top of chapter 6 on mapping camera space to XYZ, they discuss the reasoning behind supporting two different illuminants and the recommended strategy for extrapolating to other illuminants based on the user-selected color temperature. "Extrapolation" is what you want for this type of general-purpose tool, but -- obviously! -- it's not going to get you precise results.
Ben, I'm wondering if this might not be a better mousetrap. But let's back up. To build an ICC profile, you have to feed the software a rendered, output referred image don't you? It's been processed and presumably there's an application of white balance (but maybe not). This then is 'colorimetrically correct' and if so, by what metric?
I can take an ICC profile and compare what it predicts I'll get on say 500 color patches to the 500 patch reference used to build it. While it doesn't tell me close to everything I'd want to know about how that profile performs, it's a useful start to see how well this process has worked and allows me to say how 'colorimetrically close' the process was. How would someone do this with ICC camear profiles vs. DNG profiles?
I can see this being OK in a studio setup where the illuminate is the same and consistent. But if you build said profile, what happens if you move out of that environment? The idea of building a profile (in this case DNG) that allows extrapolation seems pretty useful in such situations where the illuminant is a moving target. I don't know how this does or doesn't provide 'precise results', at least if one plays with WB, tint/temp to get the image appearance as they desire.
> And this sort of fuzzy logic pervades the spec. I kid you not, the first step in ``Translating Camera Neutral Coordinates to White Balance xy Coordinates'' is, ``Guess an xy value.'' They then describe an iterative process that will, indeed, provide not-bad results in a generic tool.
Don't all 3rd party raw converters have to make a guess of the camera color space at some point? Unless Nikon or Canon as an example provide the spectral response of the chip(s), who's to say what the native color space is? My understanding is everyone has to make this assumption.
> But if you actually want to perfectly normalize white balance and exposure, the way to do it is to shoot a chart, dump the raw data to a linear 1.0 gamma UNIWB TIFF, build a matrix profile, do a reverse lookup of D50 white, and use the resulting RGB values to derive linear channel multipliers for subsequent raw development. There's no guessing and the results are perfect within the error bars of your equipment.
Moving the issue of target gamut aside (can a target define the gamut of such a device?), again, isn't this process happening on rendered data and how then does moving the camera system out of the environment where the target was shot affect these results?
> It should be obvious why Adobe wouldn't take such an approach for their tools, and equally obvious why, as a result, their tools aren't capable of delivering the same results as different tools actually intended for the job.
I'd love to see this demonstrated. Is there any source that could provide two such samples and provide a means to evaluate this?
> Sure, you can strap a canoe to the roof and you can probably jerry-rig something for a fishing boat, but that's not what the car is designed for and you'll get *much* better results using something that *is* designed for the task.
True indeed. But at this point, I'm taking baby steps in this topic where I'm pretty savvy about the differences between a boat and a car <g>. Let's assume I need to understand better the limitations of both profiling processes. Can you suggest a test process or a site that has done this to see more closely what you have described in terms of ICC and DNG camera profiles?
Andrew Rodney
http://www.digitaldog.net/
_______________________________________________
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