Canon paper profiles and BPC

[keithcooper]

Something I've asked Canon about and not had an answer... 

In Canon's printing notes, they specifically say to turn off BPC when printing from PS (with their profiles)
http://www.canon-asia.com/fineart/settings/psd.html

With my own profiles (i1Profiler) the BPC option is a useful one for some papers when I'm printing from PS, and the differences are quite clear. This goes for a lot of profiles from paper manufacturers too, given they often use i1profiler.

What is different about Canon's supplied profiles and what does this say about how they are making them or tweaking them?

I'm thinking in particular of the default ones installed with printers like the PRO-1000 (where I tested quite a range of papers during my review).

[Mark D Segal]

Hi Keith,

Very interesting post and interesting area of research. Based on all I've learned over the past while, the most likely - superficial - answer to your question is that Canon is most likely using Print Studio Pro with <No Color Correction> enabled to print the profiling targets for making their profiles. Canon recommended to me, and I've tested it, that at least on OSX this is the best way of assuring more accurate profiles ("accurate" here meaning the accuracy of the numbers for each tested colour as measured from the print relative to their corresponding reference values in the image file). It would appear for now that their recommendation re BPC is to remain consistent between the way their profiles were created and the way we use them. In the article you reference, they say that PSP uses a different colour conversion method from that of Adobe's CMM. How these two methods differ in general and why they would treat BPC differently in particular are the fundamental underlying questions. I don't have an answer, but I shall try to obtain one.

[Mark D Segal]

P.S. It's likely there may not be any answers from Canon till after the New Year.

[Doug Gray]

BPC is baked into the Canon Pro-10 profiles. It's also baked into the Canon 9500 II profiles and my Epson 9800 as well. As a result, the only thing selecting BPC can possibly do add more BPC and this would lighten shadows further. Fortunately, PS and LR seem to recognize that it is already baked in and don't do additional BPC. At least PS CC 2017. Note that the link refers to the aging PS CS6.

The canned, Canon Pro10, Pro9500 II, and Epson 9800 profiles, are not conforming ICC profiles, but in different ways. All three, wrongly, do in-profile BPC in both the Perceptual and Relative Intents. This is perfectly fine and allowed for Perceptual intent since nearly anything is allowed in Perceptual Intent. However, it should not be done in the profiles by Relative Intent. When vendors do BPC in their Relative Intent profiles it makes the profiles unusable for Absolute Intent rendering as they share the same tables. The ICC recognized that some vendors had implemented Relative Intent incorrectly and clarified how it is supposed to be done (it's a pure, colorimetric process scaled only to the media white point) in the last Version 2 spec. and this has been carried over to Version 4.

Profile Maker 5 and the subsequent, I1Profiler generate correct profiles.

BPC, as currently implemented in Photoshop CC 2017, appears to handle these well leaving the profile's BPC unchanged.  I have seen no observable affect of BPC on the canned Epson and Canon profiles in any rendering intent. OTOH, I1Profiler profiles. being ICC compliant, do not implement BPC in Relative Intent and so Photoshop BPC is necessary using an ICC compliant profile if a continuous response is desired in dark shadows. If visible areas of an image do not descend below the paper's black point, then a better print will result from deselecting BPC. Otherwise rather ugly blocking and even color shifts can occur on out of gamut colors near black.

My guess is that Epson and Canon implemented BPC in Relative intent to avoid this ugly effect if RC w/o BPC was used. The benefits of BPC are such that most encourage people to always use BPC. Epson and Canon leave no choice to the detriment of accurate Absolute Intent rendering. Likely a market force related decision. Few use Absolute Intent compared to the potentially more numerous group that might print using Relative Colorimetric w/o BPC and don't understand the implications.

Relative Intent, when the requested color is within gamut, should print that color, relative only to the white point. When it is outside the gamut, Like requesting a L=2 very dark black, it should print the closest color that the printer is capable of and it should report back via the reverse tables the L value of the actual color that will be printed. That's around 7 for the paper and setting types shown in the attached graphs.

The attached chart shows the L tone curves of  1 I1Profiler 9500II custom profile (Costco), 2 Canon canned profiles (9500II and Pro-10), and 1 Epson canned profile (9800). Both Perceptual (Blue lines)  and Relative (Orange lines) tone curves are shown for each profile. All are neutral grays from black to white. The requested L* is shown on the X axis. The value of L* returned from the profile round trip is the Y axis.

Only the I1Profiler shows the correct responses. The L value on Relative Intent clips when the requested color is below L=8 as seen by the plateaued orange line in the upper left chart. Note how the Perceptual intent show closely matching values all the way from L=0 to L=100.

It's also interesting that Epson and Canon roundtrip Relative Intent differently. Canon, at least, is properly reporting printed L levels. Epson has the curious property that a requested color printed with RI at L=0 is reported to actually print L=0. This is true even with their matte profiles that have a black point over L=20!  Totally hosed.

[keithcooper]

Thanks for that info, it confirms what I'd suspected they'd done (but had not quite got the patience to test ;-)

Do you know what they've tweaked inside the profiles to bring about this 'feature'?

[Doug Gray]

Thanks for that info, it confirms what I'd suspected they'd done (but had not quite got the patience to test ;-)

Do you know what they've tweaked inside the profiles to bring about this 'feature'?

I'm afraid I don't. I presume it was some setting on the software they created the profile with. It doesn't look like the result of profile editing. Another strange thing is that the Canon profiles for both the 9500 and Pro-10 have only 9 points separating the cubes in the AtoB tables. This is really quite small and unusual. To get away with this they need to have a much smoother response across the RGB range than printers like the 9800 where anything below RGB 120,120,120 is quite dark.

[henrikolsen]

The requested L* is shown on the X axis. The value of L* returned from the profile round trip is the Y axis.

Only the I1Profiler shows the correct responses. The L value on Relative Intent clips when the requested color is below L=8 as seen by the plateaued orange line in the upper left chart. Note how the Perceptual intent show closely matching values all the way from L=0 to L=100.

It's also interesting that Epson and Canon roundtrip Relative Intent differently. Canon, at least, is properly reporting printed L levels.

Shouldn't the Costco profile with i1Profiler also report correctly the printed L levels for perceptual rendering intent? It does 0 in, 0 out - not likely what's happening on print/paper. I would expect it to start at same black level as for relative intent. So I would say to "Only the I1Profiler shows the correct responses", that it only applies for relative intent.

Regards
Henrik

[Doug Gray]

Shouldn't the Costco profile with i1Profiler also report correctly the printed L levels for perceptual rendering intent? It does 0 in, 0 out - not likely what's happening on print/paper. I would expect it to start at same black level as for relative intent. So I would say to "Only the I1Profiler shows the correct responses", that it only applies for relative intent.

Regards
Henrik

Perceptual intent maps L=0 to the black point and the inverse Perceptual tables will map the black point back to L=0.  Perceptual intent tables are, ideally, completely inverting. This is consistent with the ICC specs.  If you want to see the actual color of the black point, one converts back using Relative Intent to scale against the white point or Abs. Col. to get the unscaled (absolute) color.

[henrikolsen]

Perceptual intent maps L=0 to the black point and the inverse Perceptual tables will map the black point back to L=0.  Perceptual intent tables are, ideally, completely inverting. This is consistent with the ICC specs.  If you want to see the actual color of the black point, one converts back using Relative Intent to scale against the white point or Abs. Col. to get the unscaled (absolute) color.

I need to understand this better it seems. I have recently been using Gamutvision's black and white density plots to evaluate profiles, especially the view "CIELAB Output L*, a*, b*, and c* (chroma) vs. input L*", and thought it could be comparable to your plots.

For my own Argyllcms profiles, my relative intent plots match the i1Profiler example you give, showing (without BPC) how black levels "give up" and flat line once max density on paper has been reached, blocking remaining shadows at this level when lower levels are input. As expected.
For perceptual intent I'm used to seeing black levels (output) start around the same max paper density, and not from 0, and I'm therefore confused.

Perhaps it's because your plots are showing the full roundtrip, and Gamutvision's is not - possibly mapping back through abs or rel as you suggest to see printed output values.

Much appreciated if you can help enlighten me.

http://www.gamutvision.com/docs/gamutvision_displays.html#BW
http://www.gamutvision.com/docs/images/Gamutvision_BW_density_LAB.gif

[Doug Gray]

I need to understand this better it seems. I have recently been using Gamutvision's black and white density plots to evaluate profiles, especially the view "CIELAB Output L*, a*, b*, and c* (chroma) vs. input L*", and thought it could be comparable to your plots.

For my own Argyllcms profiles, my relative intent plots match the i1Profiler example you give, showing (without BPC) how black levels "give up" and flat line once max density on paper has been reached, blocking remaining shadows at this level when lower levels are input. As expected.
For perceptual intent I'm used to seeing black levels (output) start around the same max paper density, and not from 0, and I'm therefore confused.

Perhaps it's because your plots are showing the full roundtrip, and Gamutvision's is not - possibly mapping back through abs or rel as you suggest to see printed output values.

Yes. It is because the round trip (blue lines) used the Perceptual tables in both directions BtoA0->AtoB0. This typically isn't even interesting because these tables are supposed to be self inverting. That is, Perc roundtrip to Perc., (the blue lines) should be just a straight lines from 0 to 100. Normally I wouldn't bother to even plot them. I chose to include these because of the odd and non-inverting behavior. Usually, what apps do, as in GamutVision,  is plot BtoA0->AtoB1 which provides the actual colorimetric response of the Perceptual transforms. Or it will if the ICC profiles follow the spec.

The canned Epson and Canon profiles demonstrate broken Rel. Col transform tables and the Canon has Perc. broken as well.

Much appreciated if you can help enlighten me.

http://www.gamutvision.com/docs/gamutvision_displays.html#BW
http://www.gamutvision.com/docs/images/Gamutvision_BW_density_LAB.gif

[henrikolsen]

Yes. It is because the round trip (blue lines) used the Perceptual tables in both directions BtoA0->AtoB0. This typically isn't even interesting because these tables are supposed to be self inverting. That is, Perc roundtrip to Perc., (the blue lines) should be just a straight lines from 0 to 100. Normally I wouldn't bother to even plot them. I chose to include these because of the odd and non-inverting behavior. Usually, what apps do, as in GamutVision,  is plot BtoA0->AtoB1 which provides the actual colorimetric response of the Perceptual transforms. Or it will if the ICC profiles follow the spec.

Got it, and see the perceptual round trip here had a special and unusual purpose to illustrate an odd point about their error.  BtoA0->AtoB1 is what I normally look it. Thanks a lot, Doug.