Pascal de Bruijn wrote:
On Fri, Jan 22, 2010 at 4:21 PM, Klaus Karcher <lists@xxxxxxxxxxxxxxxxx> wrote:Looks like the char was pretty much underexposed and your image rather overexposed.Please see my comments below."RGB(Y).pdf" shows neutral patches in your original TI3 file vs Y.What does this mean? That my camera is not completely perfectly nonlinear?
Well ... the issue is not non-linearity (your camera is fairly linear). The problem is under-exposure.
The largest RGB value in your TI3 file is approx. 80%. A standard profile will map the lightest neutral patch of your testchart (RGB approx. 80%) to Lab 100/0/0 (rel. col). RGB values above 80% will be clipped. Even when you extrapolate the data to Y=100 (e.g. with colprof -u), RGB values above approx. 85% will still get clipped (that's what I wanted to demonstrate with my PDF file).
The reason why the clipped highlights exhibit color artifacts with Lab profiles but remain more or less neutral with XYZ profiles is that you clip in different spaces along different planes. In simple terms: With a matrix profile you clip one vertex of the XYZ cube orthogonal to it's body diagonal (where the body diagonal is the gray axis). The result remains neutral as you clip the X,Y an Z component more or less equally. In Lab or Jab, you are cutting of a slice parallel to the ab-Plane, i.e. you clip only the lightness component while the color information remains untouched. You crate "zombie colors" like Lab 100 -20 0 this way and even small color variations can cause huge color shifts after clipping.
Your CMP-test.ti3 gave very similar results to my brightly exposed chart. However, the problem with both is that images with either of thoseprofiles applied are darkish and contrastless.
As your photos are more or less scene-referred, you need to apply tone mapping methods and color appearance adjustments to obtain pleasing output-referred images. See e.g. <http://www.color.org/icc_white_paper_20_digital_photography_color_management_basics.pdf>
Klaus
