I have been following another interesting thread entitled:
New Approach for Generating Optimal Profile Patch SetsIn the discussion cited above, Doug Gray presented an idea of combining two uniform grids of RGB triplets by offsetting the values of a smaller grid such that those values are packed and centered within the larger grid. Graeme Gill noted the same "body centered cubic" technique can be invoked in ArgyllCMS and also cited earlier literature where the concept has been described previously.
I’d like to share some more thoughts on what I’ve been calling “interstitially nested grid” color charts or more simply “nested grid” charts. The concept is the same as the body centered cubic approach, but I have been packing even more grids within grids. The resulting color charts are part of an ongoing collaborative Z3200 profiling project with Mark Lindquist of Lindquist Studios and John Dean of Dean Imaging. Hopefully, Mark L. and John Dean will chime in with some more comments on our findings as the thread evolves.
First, Mark L. and I started over a year ago just mastering the “art” of getting the Z3200 to print and measure custom reference charts. We retraced the steps outlined by Geraldo Garcia in this thread.
Using the HP Z3200 spectrophotometer to generate profiles for other printers. We ran into a number of technical hurdles (“bugs” more like it in the HP utility), but finally got that process down to a repeatable craft. Next, Mark L. challenged me to make a very high patch count reference chart that could be read by the Z3200. So, I took the largest patch set generated by the i1Profiler app (6000 patches), outputting the chart in CGATS format, then reformatting it using Excel and
Patchtool in order to get it into a final text format such that the Z3200 would read it. Then Mark printed and measured this 6000 patch chart on his Z3200. I ran the measurement file through BasicColor DropRGB and Mark L. also used ArgyllCMS to create a custom ICC profile. Both apps yielded very similar output using Relcol/bpc rendering from Photoshop. Mark discusses that effort here:
Amazing Breakthrough - HP Z3200ps 6000 Patch Target ICC ProfileMark L. then shared the 6000 patch reference chart and methodology for use on the Z3200 with John D. and John took it one step further. He sent Mark and Mark each a set of prints made with various B&W printing methods on different printers plus two color prints. I will let Mark L and John discuss that part of the project if they wish to, but we all concluded our Z3200’s produced prints with subtle but real improvements in smoothness, color discrimination, and grayscale neutrality with the 6000 patch profiles compared to profiles built with the classic 12x12x12 (1728 patches) uniform grid chart design. FWIW, buried inside the i1Profiler 6000 patch target is an 18x18x18 (5832 patches) uniform grid plus 168 more patches chosen by the i1Profiler app with some proprietary algorithm to round out the patch count to 6000!
Suffice to say, Mark L. then challenged me to go bigger and bolder than 6000 patches! And after a hiatus caused by other project commitments, I finally took up that challenge last November. I built a number of custom nested grid targets, the biggest one being 10445 patches! And we’ve been printing, measuring, and looking at output from them. I have attached a couple of photos of a nested grid chart design having a total of 3315 patches. The Z3200 automatically lays out the color chart patches on the fly depending on what size media one chooses (roll or cut sheet). A Patchtool‑scrambled version of this 3315 reference chart is shown laid out to print on a 42 inch roll. The other photo shows the 3315 nested grid chart laid out for printing on a 24 inch roll. Because the patches for the 24 inch wide print were sent without randomization or scrambling in the reference chart file, the nested grids can be clearly observed in the chart pattern.
The 3315 target shown here consists of 1) a 9x9x9 (729 patches) uniform grid, followed by 2) an 8x8x8 (512 patches) offset so as to fit centered within the 9x9x9 grid, followed by 3) a 16x16x16 offset 8 grid and filtered on HSB saturation for those colors occupying 22% of 9x9x9 RGB cube volume (644 patches), thus centered about the diagonal neutral line in the primary RGB cube and finally, 4) a 32x32x32 offset 4 interstitial grid filtered on HSB saturation for those colors occupying 11% of RGB cube volume (1428 patches) centered about the diagonal neutral line in the RGB cube.
So what does this interstitially nested grid concept do?
If you fill the primary grid with correctly offset values derived from a grid just one size smaller, it essentially gives you the precision of a much larger uniform grid without using as many patches. Indeed, it essentially becomes one much larger uniform grid everywhere except at the facets of the RGB cube, thus dropping the precision back to the same precision of the primary grid but only on the facets of the RGB cube. I tend to think of it as a “lossy compression” technique for color charts where the values thrown away on the facets of the main RGB cube reduce the chart size significantly but only at locations which don’t require so much spacing precision. For example, combining a 9x9x9 grid with an interstitially centered 8x8x8 grid = 9+8 = 17, i.e., the equivalent of a 17x17x17 uniform grid everywhere except on the facets of the RGB cube. The facets drop back to 9x9x9 spacing precision. Because the colors on the edges and facets of the cube represent the most saturated colors possible at any given L value generated by the printer as long as the printer isn’t seriously over inking the media we are sacrificing precision only where we really don’t need it.
My research on the I* metric taught me that observers can accept much less color accuracy in vivid colors compared to low chroma colors and the I* is weighted to take that into account (much more than dE2000 weights this response). Hence, the practical benefit of a nested grid color chart is that it dramatically boosts the RGB spacing precision where we do want it while helping to keep total patch count down to a more reasonable amount. For example, in a 9+8 nested grid design, inserting the 8x8x8 grid into a 9x9x9 grid combines 512 patches with 729 for a total patch count of 1241. This nested grid chart punches well above its weight because it is behaving identically to a17x17x17 uniform grid over approximately 94% of the entire RGB cube volume. If one chose a single 17x17x17 inch uniform grid for a color chart, it would contain 4913 total patches! That’s just one example of the practical value of the nested grid or “body centered cubic” method as Graeme Gill likes to call it.
A second example of patch count efficiency is found in our 10445 patch chart. I kid Mark L. that he has inspired “The Mother of all profile charts”! Totally outrageous but doable with the Z3200 spectrophotometer. It was designed with monochrome printing in full color mode in mind, and it maps 94% of the RGB cube with 33x33x33 equivalent (i.e., 17 + 16 = 33) uniform grid spacing, but it also maps the low saturation (neutral and near neutral) RGB triplets with 4 RGB unit spacing precision that would otherwise require a single 65x65x65 uniform grid. A 65>3 uniform grid would map 100% of the RGB cube at 4 unit spacing intervals, but it would also contain 274,265 patches! Hence, our 10445 chart was a good test as to whether the Z3200 can be overfitted with too much spacing precision which might cause profile accuracy and smoothness to decline. We have found that it does not cause a decline in quality, and in fact the quality of the output is superb. How much better is the print quality than, say a 2000, 4000, or 6000 patch chart? We are working on this question now. The differences are indeed subtle, but one thing we have proven for sure. The Z3200 prints with noticeably higher print quality when patch counts climb above the biggest chart HP supplies with its Color utility, i.e., a 1728 12>3 RGB cube color chart. So custom charts with extended patch counts are indeed worth pursuing.
Another advantage of nested grid chart designs, IMHO, is that they can place more precision into lower saturation colors where it is often beneficial without having to resort to a two step optimization process or know precisely where the neutral gray scale is occurring. I’d rather approach my profiling activities with a “one and done” target printing session!
Mark, John, and I are now working on both objective and subjective evaluations comparing various nested chart designs to other conventional charts with both smaller and greater total patch counts. As I noted earlier, we quickly concluded that our Z3200 printers really do benefit from charts greater than the classic Atkinson 1728 patch target. That said, where the sweet spot lies is going to take more testing. Moreover, as I study how ICC profiles are evaluated these days both objectively and subjectively, the current methods may very well need to be augmented with yet unforeseen approaches to color metrics and to printer evaluation targets.
Enough for now. I hope Mark Lindquist, John Dean, and others will also share their thoughts in this thread.
Cheers,
Mark McCormick-Goodhart
http://www.aardenburg-imaging.com