2007.10.24 16:47 "Re: [Tiff] Floating-point 32bit sample data", by William B Thompson
On Tuesday 23 October 2007 01:09, Nikita V. Borodikhin wrote:
Are floating point RGB data stored gamma corrected or linearized? How should I use ICC workflow while processing data? ICC workflow is based on integer data, not float...
On Tuesday 23 October 2007 15:38, Chris Cox wrote: > HDR data _really_ needs to be gamma 1.0, and floating point image data is > best represented as gamma 1.0.
Photoshop CS3 appears to exhibit the following behavior with 32 bit(float)/channel images:
- An 8 bit/channel image without an attached color profile, read into Photoshop and then converted to 32 bit(float)/channel, is converted to linear gamma using the default color space. If this image is saved as a 32 bit(float)/channel TIF, it gets an attached profile with the name of the default color space and "(Linear RGB Profile)" added to the end. Gamma values are set to 1.0 in this profile (or possible they are missing?).
- An 8 bit/channel image with an attached color profile, read into Photoshop and then converted to 32 bit(float)/channel, is converted to linear gamma using the attached color color profile. If this image is saved as a 32 bit(float)/channel TIF, it gets and attached profile with the name of the original color profile with "(Linear RGB Profile)" added to the end. Gamma values are set to 1.0 in this profile (or possible they are missing?).
- A LogLuv TIF image read by Photoshop then written back out as a RGB floating point TIF is assigned the "XYZ Profile", which specifies linear gamma and an identity transform. It appears that the RGB channels in fact have XYZ values, which is what is also done in EXR. Photoshop CS3 cannot write LogLuv TIF images.
The "(Linear RGB Profile)" versions of the ICC profiles do not appear in Program Files\Common Files\Adobe\Color\Profiles in Windows XP, nor does the "XYZ Profile".
The document describing the Adobe RGB (1998) Color Image Encoding specifies that (0.0, 0.0, 0.0) should correspond to the color space back point and (1.0, 1.0, 1.0) should correspond to the color space white point. It further specifies that "Component values outside the range [0, 1] are not allowed for floating-point encodings."
Photoshop CS3 allows assigning profiles to 32 bit(float)/channel images, but does not support profile conversion (which would have the effect of rescaling the image). It reads and writes 32 bit(float)/channel TIFF images containing values outside this range without clipping.
- Bill
--
William B. Thompson
Professor, School of Computing, University of Utah 50 So. Central Campus Dr., Rm 3190, Salt Lake City, UT 84112
thompson@cs.utah.edu, 801-585-3302 (ph), 801-581-5843 (fax) http://www.cs.utah.edu/~thompson