Skip to content

dev_nchannel_profcheck.md

Latest commit

 

History

History
56 lines (46 loc) · 3.09 KB

dev_nchannel_profcheck.md

File metadata and controls

56 lines (46 loc) · 3.09 KB
Raw

N-channel profile accuracy check (profcheck for CMY+N)

ArgyllCMS's profcheck refuses device colourspaces beyond 4 inks — its .ti3 reader is hardcoded to Grey/RGB/CMY/CMYK (the "generic colorant read" in its TTBD is unimplemented; pval.p[] is filled from only ci/mi/yi/ki). So Check & Refine used to hard-error on a 6-ink profile with "use RGB, CMYK or another standard colour space".

workflow/profcheck_nchannel.py reproduces profcheck for any channel count in Python, driving Argyll's icclu for the one thing that needs the profile:

  1. Read the .ti3 generically — device fields are <REP>_<letter> per ink (e.g. CMYKOG_C … CMYKOG_G); the fixed-name ≤4 reps (iRGB → RGB_R/G/B, K → GRAY_K, …) are mapped explicitly.
  2. Predict — device → PCS through the profile's forward (A2B) table via icclu -ff (icclib handles 2..15 channels), at profcheck's intent (default absolute).
  3. Measure — the file's XYZ→Lab (Argyll icmD50 = 96.420288, 100, 82.490540), or a spectral recompute (below).
  4. ΔE — CIE76 / CIE94 / CIEDE2000, computed here.
  5. Emit profcheck's exact text: No of test patches = N, per-patch [dE] SID[ @ LOC]: <dev 8dp> -> <predLab> should be <measLab>, and Profile check complete, errors[ (CIE94|CIEDE2000)]: max. = , avg. = , RMS = .

Because the output is byte-for-byte profcheck format, the whole workflow/profcheck_runner.py pipeline — summary parse, per-patch parse, quality grade, refine-strip flagging (keyed off SAMPLE_LOC, so it's ink-count-independent) — and both the Guided and Manual Check & Refine flows work unchanged for 6-ink profiles. No new compiled binary (unlike the gamut helper — icclu is enough).

Routing

ProfcheckRunner.run counts device channels (_ti3_device_channels): >4 runs the Python check in a _NChannelWorker(QThread) streaming the same lines; ≤4 still uses the stock profcheck binary. So nothing changes for RGB/CMYK.

Spectral options (-i illuminant / -o observer / -f FWA)

profcheck can recompute the patch colours from the raw spectral data before comparing. Argyll's own spec2cie does exactly that — but it also refuses

4 ink. Since it only integrates the spectral columns, _recompute_spectral_xyz hands it a relabelled RGB copy: dummy RGB device set to each patch's brightness (so FWA can find the media-white patch), the real SPEC_* columns, and all original header keywords (TARGET_INSTRUMENT etc. — FWA needs the instrument illuminant). It then merges the recomputed XYZ back by SAMPLE_ID. Result: bit-exact Argyll colorimetry + FWA, no reimplementation. If the .ti3 has no spectral data, it errors cleanly rather than silently using D50.

Validation

Line-for-line against real profcheck on RGB/CMYK (where both run): summary max/avg/RMS identical and predicted Lab identical; only a 1e-6 rounding differs in the ΔE display. -i D65, -o 1964_10 match to 1e-6; -f delegates to spec2cie, so it errors identically when the data carries no FWA illuminant. Tests: tests/test_profcheck_nchannel.py.