path: root/ttbd.txt

Argyll - Things To Be Done & odeas (in no particular priority)
--------------------------

Categories:

1)  Ideas list
2)  General
3)  Documentation
4)  Charts
5)  Readings/Instruments
6)  Profiles
6a) icclib
7)  Linking, gamut mapping
7a) Profile/Link speedups
8)  Raster conversion
9)  Scanin
10) Calibration
11) Bugs

===================================================================
Recent suggestions:

	Add troublshooting section to tutorial:
		colprof report fit DE's	
		iccgamut
		transplot
		xicclu fwd & bwd and both
		cctiff through test chart path
		cctiff soft proof

	Add flat field support to scanin

	Add DICOM support to dispcal.

	Change CIECAM02 to simplified & improved version

	add filtered mode for tiffgamut (done ?)

	add dispcal option to set white point to closest point
	on the white locus to RBB 1 1 1

	add option to dark cal spectro's every N readings in dispsup.

	Add some i1pro RevE UV support for improved FWA

	Add "single pixel chart" mode to printtarg and scanin.

--------------------------------------------------------
1) List 2


1a) Candidate list:

* Get optimised separations working for CMY->CMYK & smoothed sep. option

* Implement ICC V4

* Add monochrome suppport

* Fix plot to have seperate thread etc, and
  add "save spectrum support" in spotread.

* Fix xmatrix.c to allow for curve offset - this will
  improve quality of non-calibrated display matrix profiles greatly.


2) General:

* Add full support for use as DLL/SO libraries

* Check all executables & libraries with valgrind, and
  clean up leaks and unitialised memory.

* Merge changes from Rhett Sutphin into iccdump that support
  dumping embeded profiles from known file formats.

* Create a utility for converting color profiles to monochrome.
  ie. CMYK -> K only, RGB -> R=G=B profile.
  Have option for forcing the profile to be a color profile
  but monochrome output.
  Have an option for outputing a clut based or ICC monochrome profile.

* Add support for ink-subset profiling, e.g.
  Monochrome like inks only (light + K etc.).
  For full support, would have to support gamut mapping
  of 2D colorspaces.

* Add support for an Argyll private ICC tag, and use it to
  convey the device ink limit, and other profile creation
  information for improved linking checking and convenience.

* Add support for a (private tag) real gamut clut as a 3D
  surface in profile.

* Add > 4 color device optimised separations creation support.

* Add optimised separations handling for CMYK profile generation option.

* Switch to balltree acceleration for gamut and rspl rev code.

* Switch to ball tree for targen spatial acceleration ??

* Re-code all of argyll to be a consistent set of APIs,
  to make it a full "library". Python interface ?

* Fixup error and allocation code in all the Argyll libraries.
  Add a better error reporting mechanism ?

* Write utilities to convert ICC profiles into PostScript CSD and
  CRD's.

* Define an open operating system color service API, and reate
  an example implementation with Argyll.

* Make printspot more flexible - load spectrums, compare them
  etc. - similar to Xrites DTP41 utility.

* Create a device "linearity" measure utlity. Read in a CLUT
  profile and compute the first and second differences of the grid.
  Report grid res. scaled average & maximums ?
  (What about input and output curve effects ? - sample overall
   chacateristics ??)

3) Documentation:

* Add page on viewing condition parameters and typical values,
  to link to from colprof, collink etc.

* Add help in generating greyscale test charts into targen doco.(-f0 etc.)

* Add doco on intependent verification of profile accuracy.

* Document icclib api (Do this with V4 release!)

* Document cgatslib api

* Add "advanced scenarios" section to Argyll, to explain custom lighting,
  FWA, proofing fine tuning, etc.



4) Charts:

* Fix i1pro chart to allow for arbitrary long paper.

* Create a test chart specifically for establishing an ink limit ?

* Should create a .ti2 template file for some standard charts,
  such as an IT8.7/3, ECI2002 random and non-random etc.
  Scanner recognition files too ??

* Add an option to targen, that allows generation of
  test points down the neutral axis (how does this work
  for CMYK, since there are many possible values ?? -
  use ppoint with 1D target and 2D device ??)
  - try enhancing current CIE94 effect ??
  + other memory colors like skin tones etc.
  
* Started work on tiffsamp.c, but not complete.
  Utility for creating representativ test points
  from sample device images.

* Create tool workflow to allow creating a "worst case"
  verification chart based on the veronoi locations of
  a chart used to create the profile needing verifying.


5) Readings & Instruments:

* In chartread, add RGB and Monochrome "generic"
  profiles, so that bad readings can be recognised (??)

* Create utility that translates between an XYZ only .ti3 and
  an estimated spectral .ti3, based on a spectral .mpp,
  to allow applying WFA compensation using non-spectral
  instruments. (is this technically feasible ?)

6) Profiles:

* Determine typical statistical distribution of errors
  in relation to Lab space, and use per sample point
  weightings to apply more uncertaintly specific smoothing 
  factors.

* Add option to use XYZ as PCS, and matrix in B2A table
  to improve efficiency of B2A conversion (hence acuracy).

* Investigate why using a black ink limit slows inversion down so much,
  and see if the performance can be improved. Shift black ink limit
  to be a calibration function ??

* Add support in ICC profile for embedding the .ti3 CGATS data within
  the profile. Add support for extracting it again in profile.

* Add a custom tag that can encode a spectral A2B profile,
  to support illunimant specification at link time ?

* Add xicc support for creating monochrome profiles from
  raw data.

* Add support for regenerating perceptual and saturation
  B2A tables in revfix. Rename utility ?

* Can FWA compensation be improved in accuracy ? It seems to
  over estimate FWA effect at high density of inks.
  Is this due to scattering effects, asymetry in the filtering
  effects of the colorant, or due to errors in estimating the
  colorants filtering effects into the UV region ?

* Modify rspl thin plate spines to have a "simplex" data
  interpolation mode, rather than its "n-linear" one, for better
  correspondence with how data is interpolated within the grid ?
  (Is this really worth it ?)

* Add smoothing for K locus generation for transtion to
  ink limiting. Need to detect region near ink limit,
  and then generate non-limited and limited values,
  and then regenerate reverse value with blended K locus
  target. Probably not important since ink limiting region
  is often very dark, and optimised separations could be used as
  an alternative mechanism.
  [ Use optimized separations to fix this. ]

* Add an option for input (and display ??) profiles to
  include both matrix and Lut tables by default. (What PCS
  should it use, XYZ for the matrix, or Lab for the
  best quality Lut ?? What other CMS need such things ?) 

* Modify the rspl simplex code (and possibly imdi code)
  to orient the simplexes in each cell to minimse the
  expected curvature error at the center of each cell.
  [ Partially implemented, but not complete.
    Optimized separations is more important. ]


6a) Icclib:

* Re-architect icclib to allow for unknown datatypes,
  dynamic custom tags, and full ICC V4 support.
  [ In progress. icclib now supports unknown data types,
    after back porting code from icclib4.]

* Add full spot color library support

* Change iccdump of Luts to dump PCS values as well as
  normalised integer values (table entries and indexes ?)


7) Linking, gamut mapping:

* Add optimized per channel input/output curves
  generation, rather than relying on the per channel
  curves provided by the profiles (ie. xicc.c approach).

* Would be nice to be able to weigh L errors more in
  gamut clipping. Would have to introduce scaling factors
  in rspl values to fake this without changing the
  insides of RSPL.

* Fix xicc library to handle matrix/shaper profile inversion
  correctly, when the target is out of gamut. Don't clip in
  device space, clip to nearest in CAM space ! (Fix after
  iccV4 changes ?)

* Fix xicc library to do better clipping for inputs that
  are out of range of the core ICC lookup mechanisms. 

* Add flag to work with "preserve black" that forces K only
  input to map to pure K only output, to suite use on press
  re-targeting. How about creating a gamut mapping to do
  this for arbatrary colorants ? - Can use "weak default mapping"
  to help curve shape ? How to handle resulting gamut shape change ??

* Add proper support for a "filtering" mode of black generation,
  that smooths the black colorant target to take into account
  gamut boundary issues. ie. revisit auxsmth.c ? Use optimised
  separations mechansim instead ??

* Add support in collink for linking with monochrome in and out
  profiles. Gamut mapping needs to have monochrome bypass of
  3D mapping.

* Add support for setting arbitrary colorant constraints
  (ie. min/max/target, sum combinations)
  This is linked with the ideas of culling the gamut to
  avoid CMYK+ aliasing problems.

7a) Profile/Link speedups

* Investigate why rev_nnfind() is so slow. Can it be
  improved ? 

* Add adaptive sampling reverse interpolation mode, to speedup
  high res inverse tables. Look at approach Done Bone used
  in adaptively measuring samples.

* Add support in collink and colprof, for multi-threading
  the reverse interpolation calculations. This will allow
  great speedups when more than one processor is available.
  Simplest approach is to have multiple instances of rspl,
  and partition the problem into separate spaces (not interleaved).
  Cache contents shouldn't duplicate then.
  Multi-threading within rspl/rev would be complicated, and
  may be slow due to contentions.


8) Raster conversion:

* Add support in IMDI for an auxiliary "calibration" or
  "linearisation" table as a parameter to the setup.
  [Calibration creation should allow for arbitrary curve
  manipulation, such as limiting maximums (to replace
  max K function), and other arbitrary curves.]

* Add optimized per channel input/output curves
  generation, rather than relying on the per channel
  curves provided by the profiles (ie. xicc.c approach).

9) Scanin

* Add check in scanin that prints warning message if
  patch variance is too high (probable faulty read).

* Add spectral fix options to scanin code to allow compensation
  for scanner and media errors when using a scanned image to
  measure color.
  This means figuring out how it will work, as well as creating
  to tools to create the spectral fix data (or just add general
  tweak/abstract profile support ?).

* Modify scanin reference .cht format to be more robust (ie. CGATS)


10) Bugs

* filmtarg  doesn't work any more with latest .ti1 data format.
  Need to fix it.