ArgyllCMS printer profile GUI is on it's way

[crenedecotret]

Hi everyone

Just thought I would drop this here... it's not entirely done but when it is I will provide instructions. This is on LINUX but python is very portable
The "generate" and "profile" portions are not done.. those are the easiest to do. Having the software read the row layout correctly and having it talk to chartread was difficult.

I also did a fun compare screen sort of like i1profiler and spyderprint but it only works once all the patches have been read. It may help spotting bad reads

 

[pharmacist]

A free version similar to i1Profiler interface for ArgyllCMS would be nice.

 

[crenedecotret]

A free version similar to i1Profiler interface for ArgyllCMS would be nice.

I doubt it will ever be the same. Argyllcms was never really intended to be used outside of the command line. I1Profiler was built for a GUI

An example is the "split" color patches to check for errors. Chartread has no way of showing us the values for the strips. It keeps everything in memory so that's why I can only display it at end.

An another almost comical example.. My program is built to be able to click a row and read it again. The way I have to do it is have Python send F and B repeatedly to chartread to reach the column the user clicks.

I'm surprised at this point the wrapper works as well as it does

I'll probably reuse some code from pkpatches for the patch generator when I'm done with the chart reader.

I'll probably never add printing routines since I want to keep this portable between Linux and Windows.

Since a GUI would probably interest beginners, I also have some ideas for a two step process (like i1studio). I'm thinking step 1, print a 98 patch colormunki format profile Step 2. Use the first profile to precondition 98 more patches. Step 3. Combine both ti3 files and profile. I due time I suppose.. I've never even tested such a process it's super hypothetical in my head right now.

 

[pharmacist]

The idea of a 2 step-method with 98 patch start target + 98 patch secondary target like the 50 + 50 Colormunki/i1Studio method would be nice and probably produce better profiles than 50 + 50 patches as more sample points are used to sample the RGB color space.

 

[crenedecotret]

I believe I was able to figure out the two step process or at very least do someting useful.

Here is a comparaison between my profiling tool (with new fancy 2 step process) and a multi-page profile I did with argyllcms

Now the first thing you are going to say is that it looks bad, but there is a caveat. The original profile was done with 1144 patches using proper procedures on an i1pro. The "combined measurements" profile was done with 30 patches on step 1 followed by 30 patches on step 2. All on 4x6 paper. This 2-step profile was done rapidly without even allowing the ink to dry for than 2 minutes. This was with a colormunki and it only took a minute to scan each set. With such a small set, I suspect that the colormunki might be better but even then, i'm not so sure. The reported gamut is smaller but it's dosent mean it's not accurate. It's probably OK for cheap snapshots on 4x6 paper. I havent printed with it yet. There is a difference in the soft proof as one would expect (remember 60 vs 1144 patches)

I'm hoping that with a 1 page + 1 page approach on letter or A4 paper where we have a realistic number of patches, we might get a better result than a normal two-page profile

Here is the method... the first page uses targen and then generates a preliminary profile. It can be small and super easy to measure with something like a colormunki (I'm allowing single and double density modes)

For the second page, I'm using code that generates a custom ti1. The code builds a candidate pool of about 10000 rgb values using a few strategies: uniform 3d grid, mathematical space filling, neutral axis and hardcoded perceptual anchors (skin tones, foliage, sky, etc).

Each of the candidates is then converted to lab space through xicclu and then graded based on
- novelty: DE distance from the nearest existing patch in pass 1, but i soft cap at DE 30 to avoid having candidate patches that the printer will never be able to improve
- region undecoverage: the algorhytm will add weight to areas we havent sampled as well
- neutrality bonus: since the 2nd set is an optimisation set, we give some additional weight to near neutral colors
- luminance balance: patches in under-represented luminance areas (shadows/mids/highlights) are given some extra attention

Once we have our graded candidates, we add one white, one black and 3 gray patches. The remaining patches are then selected based on their ranking BUT they need to be at least 2.5 DE away from the patches in the original step 1 measurements.

The result is a second patch set that measures where the preliminary profile was weak and was missing information. At least that's the theory

At the end, I combine the ti3 measurements from step #1 and step #2 and hope that colprof now has enough information to build a decent profile.

I'm probably run a bigger next over the weekend with full colormunki double density pages

 

[crenedecotret]

I couldnt resist and I did a test print.

For 60 patches TOTAL, it's not terrible. For color images, I'm seeing a very small color cast towards yellow but it's only noticeable side by side with the 1144 patch profile. Black and white is pretty much unuseable with this profile.

This was matte paper loaded with OBA which is already a tiny bit more difficult to get good results due to the limited bandwidth. The small patch count might get better results with a better ink/paper combo but I ended up marking the 4x6 selection as experimental in the software.

The ink/printer i'm using is already pretty hard to profile.