Modeling of tone reproduction curve
I made an approximate model of tone reproduction curve by regression analysis on the color data from the scanned target image. The model has discrepancy to some degree from the color data, because the model isn’t interpolation. The ICC profile will be calculated from the model. So the accuracy of the ICC profile is limited to the model. If the discrepancy is quite small, the ICC profile can be quite accurate (according to your scanner, at least). If the discrepancy is large, there is no hope.
The histogram above shows the distribution of the discrepancy by color difference ΔE00. You cannot find ΔE00 < 1 difference by your eyes. You feel ΔE00 > 4 as noticeable error. For your information, the color resolution of 8-bit sRGB is 1.18 at worst case, (33, 33, 33) to (33, 32, 33). If you set up your printer and scanner correctly, 95% of patches becomes ΔE00 < 2. If not, review your settings.
You might need accuracy for particular area. I list worst five cases below.
| Read color | Model color | ΔE00 | ||||
|---|---|---|---|---|---|---|
| L* | a* | b* | L* | a* | b* | |
| foo_l | foo_a | foo_b | bar_l | bar_a | bar_b | baz |
The colors are CIE L*a*b*.
Upload your test image
If you want to evaluate the ICC profile by your test image, upload it. It should be RGB with embedded profile.
Massive calculation
I need a half hour to calculate the ICC profile. Click the next button below, and have a walk. I’ll notify you by email after it is complete..