A detailed comparison of I1Profiler and Profile Maker 5

[Doug Gray]

Thought I would compare various aspects of the profiles generated by PM5 and I1P. In particular two attributes are the most important to me. Print color accuracy and color reporting accuracy. I did not look at the way the two handle Perceptual Intent printing as that is both a matter of taste and both programs have various, ill described  settings that influence Perceptual Intent.

Actual print color accuracy is simply the ability to request printing a Lab color and get reasonably close. For reproduction work such as duplicating old documents or images, maintaining color control close enough that the documents and print visually match is the basic requirement. In my experience getting most dE's below 2 is sufficient in most cases. Also, getting the image wide average dE, below 1 is as important so the two don't differ noticeably from each other. This is a function of the B2A1 interpolation accuracy, instrument and printer error and noise.

The other factor important to me is the accuracy of the reverse color lookup, or the A2B1 interpolation process. This should be as quite good as it only involves interpolation errors. Interpolation error is influenced by the lumpiness of the printer RGB space. The more 3D grid points, the closer the programmers can get the A2B1 table to accurately invert the B2A1 table. While the A2B1 tables are not used for printing they are used for proofing and cross rendering.

The test used the following setup:
The printer used is a Canon 9500 II which I've found provides reasonably consistent results and its RGB response is smooth enough that it doesn't benefit significantly from very high patch counts.

Profiles for both the I1Profiler and Profile Maker 5 were created using a standard 918 chart with 25 stepped neutrals added to enhance B&W uses. The charts were scanned twice with an I1Pro 2 and averaged. The spectral results imported to PM5.

I1Profiler's settings were default except that the large profile table was selected. This produces a 37x37x37x3 Lut. Profile Maker 5's settings were "large" and Logo Chromo Plus. These settings produce a 33x33x33x3 Lut.

Then I randomly selected points in the printer's RGB space. 150 were reversed using the PM5 profile and the other half with the I1Profiler profile. These Lab values were printed to make a I1Pro 2 scan formatted image.

I was expecting I1Profiler to slightly outperform Profile Maker 5 because of the higher number of Lut grid points. This turned out to be the case but the surprising thing is how little the the Lab colors changed during round tripping from Lab to printer to Lab. For the I1Profiler the average change was dE2k .21 and for ProfileMaker the change average was .56.

The I1 Profiler also proved to provide slightly better physical print accuracy coming in at an average dE2k of .60 v .83 for ProfileMaker 5.



The formatting is irregular because the BB fonts aren't monospace.

I1Profiler Reports:
ICC Profile Reported Worst 2% Colors
Req:  34.76  -3.30 -57.26     Act:  35.24  -5.35 -57.61     dE2k: 1.33
Req:  27.77   8.40 -56.29     Act:  27.74   6.84 -56.26     dE2k: 1.02
Req:  31.74 -24.57 -37.74     Act:  32.32 -26.22 -37.30     dE2k: 0.94
Req:  76.01   3.27   9.51     Act:  75.78   2.73  10.06     dE2k: 0.89
Req:  10.19   9.53  -9.85     Act:  11.56   9.38  -9.80     dE2k: 0.87
Req:  31.79 -17.49 -41.51     Act:  32.10 -19.13 -41.79     dE2k: 0.87
Requested v Reported Labs:  Max:1.33,  Ave:0.21,     Best 90%:   Max:0.48,  Ave:0.16   Shift:[-0.05 0.09 -0.02]

Printed Worst 2% Colors
Req:  51.42 -23.68  52.68     Act:  54.31 -24.27  54.73     dE2k: 2.88
Req:  48.29   4.64  49.89     Act:  50.36   3.41  51.88     dE2k: 2.35
Req:  75.93  16.07  86.31     Act:  77.64  12.93  86.52     dE2k: 2.16
Req:  19.59  -8.05 -17.45     Act:  19.77 -10.57 -17.34     dE2k: 2.08
Req:  55.82   2.20  58.43     Act:  57.70   0.82  59.61     dE2k: 1.99
Req:  69.18  27.11  81.05     Act:  70.93  24.33  80.24     dE2k: 1.96
Requested v Printed Labs:   Max:2.88,  Ave:0.60,     Best 90%:   Max:1.03,  Ave:0.50   Shift:[-0.07 -0.07 0.19]


ProfMaker 5 Reports:
ICC Profile Reported Worst 2% Colors
Req:  32.49  40.14  27.61     Act:  32.13  39.93  19.91     dE2k: 4.39
Req:  35.37  21.30  35.32     Act:  35.19  22.43  27.82     dE2k: 4.33
Req:  45.13  56.89  52.23     Act:  45.42  59.05  45.53     dE2k: 3.67
Req:  43.48  66.48  54.40     Act:  43.93  69.04  47.93     dE2k: 3.51
Req:  16.44   8.71   5.17     Act:  16.36   5.41   4.17     dE2k: 3.37
Req:  31.31  -6.16  20.65     Act:  31.08  -6.29  16.38     dE2k: 2.49
Requested v Reported Labs:  Max:0.08,  Ave:0.56,     Best 90%:   Max:1.13,  Ave:0.40   Shift:[-0.04 0.19 0.31]

Printed Worst 2% Colors
Req:  16.44   8.71   5.17     Act:  16.01   5.56   4.33     dE2k: 3.21
Req:  75.93  16.07  86.31     Act:  78.33  11.22  86.27     dE2k: 3.21
Req:  19.59  -8.05 -17.45     Act:  19.64 -11.43 -16.10     dE2k: 3.03
Req:  43.48  66.48  54.40     Act:  44.31  68.72  49.67     dE2k: 2.77
Req:  15.98  10.51 -23.40     Act:  15.53   8.85 -25.62     dE2k: 2.58
Req:  69.18  27.11  81.05     Act:  70.91  23.34  80.79     dE2k: 2.44
Requested v Printed Labs:   Max:3.21,  Ave:0.83,     Best 90%:   Max:1.57,  Ave:0.68   Shift:[-0.05 0.51 0.19]

[Erland]

Thanks for sharing! I also had a hunch i1P would make a better result.
I sometimes use PM5 only because it produce a better dmax with some ink-sets. Though I often its because PM5 allow a slighty more non-neutral black (like if you use a 3rd party ink-set).

[Doug Gray]

Thanks for sharing! I also had a hunch i1P would make a better result.
I sometimes use PM5 only because it produce a better dmax with some ink-sets. Though I often its because PM5 allow a slighty more non-neutral black (like if you use a 3rd party ink-set).

You're welcome.

As for Dmax, I haven't seen any difference on the RelCol side but I would expect the Perc. Intent would be different.

EtoA:
Indeed, the default PI is quite different between PM5 and I1P, the latter has more contrast. This creates the perception that the media BP is lower. I also prefer the I1P PI for normal printing where I'm not trying to accurately portray color.

One of the things I didn't analyze in the initial post is the differences between the actual print and reverse rendered image. It turns out I1Profiler is much closer than PM5 as well. The former has only 3 (1%) patches that are greater than dE2k of 2 while PM5 has 19 (6%) of the patches greater than 2 dE2k.

This particular metric bears on the accuracy of soft proofing.

Even though the I1 beats PM5 significantly, I can't see any differences for in-gamut printing using Rel Col. The difference is more critical when doing something like precision printing Logos where color accuracy and matching is given close scrutiny.

EtoA:
I found this comment in Aryll's notes on color profile ambiguities. It particularly applies to PM5. Interesting that the latest V4.3 ICC spec emphasizes invertability in the A2B LUts.

An alternate approach to solving these problems is to ignore the BToA LUTs, and use a reverse lookup of the AToB LUT. Working against this is the fact that many profile creators, creating profiles for output devices, seem to store the AToB LUTs with reduced resolution,  assuming the the BToA LUTs are of primary concern.

Attached is a histogram of the two:

[GWGill]

The other factor important to me is the accuracy of the reverse color lookup, or the A2B1 interpolation process.

I would call the A2B the "forward" table for an output device, since it corresponds to the natural operation of it :- device values in, color out.

The more 3D grid points, the closer the programmers can get the A2B1 table to accurately invert the B2A1 table.

The B2A is typically the inverse of the A2B, since it is the A2B that represents the actual behavior of the output device. The B2A is used to predict what device colors will produce the desired output color.

For a 3 channel output device it is easy to cheat a B2A by swapping the raw in/out data and interpolating (and it is even possibly to do this for CMYK with a bit of effort), but you end up with some slightly different tradeoffs in terms of accuracy and knowing what is in or out of gamut. 

[Doug Gray]

I would call the A2B the "forward" table for an output device, since it corresponds to the natural operation of it :- device values in, color out.The B2A is typically the inverse of the A2B, since it is the A2B that represents the actual behavior of the output device. The B2A is used to predict what device colors will produce the desired output color.

It certainly depends on one's perspective. From the PCS it's a reverse map, from the device it's a forward map. That's why I further identified it as the A2B1.  It removes the ambiguity.

For a 3 channel output device it is easy to cheat a B2A by swapping the raw in/out data and interpolating (and it is even possibly to do this for CMYK with a bit of effort), but you end up with some slightly different tradeoffs in terms of accuracy and knowing what is in or out of gamut.

The current ICC spec makes a point that the A2B mapping should ideally be just an inverse of the B2A for all in gamut colors within the limitations of the back and forth interpolation. Over the printable gamut the PCS uses a much smaller number of B2A LUTs whereas, typically, all the A2B LUTs are utilized. As a result one should expect B2A interpolation errors would be greater than the A2B ones.  This is certainly the case with I1Profiler though not with the much older Profile Maker 5.

And the increase from 32x32x32x3 LUTs to 36x36,x36x3 helps as well.

Oddly, Profile Maker 5 produces profiles that map LUT Lab(100,0,0) and not (100.3,0,0) to 65535 but map (a,b=0) to 16368 per the older interpretation of V2. Sort of half baked.

It seems the Photoshop and Little CMS handle the differences and correctly interpret these variations.

[GWGill]

Oddly, Profile Maker 5 produces profiles that map LUT Lab(100,0,0) and not (100.3,0,0) to 65535 but map (a,b=0) to 16368 per the older interpretation of V2. Sort of half baked.

V2 PCS encoding hasn't changed (i.e. there is no "old" or "new").

V4 uses a mixed PCS encoding - a new one for some tags, the V2 one for other older tags. (it's a mess).
[/quote]

[Doug Gray]

V2 PCS encoding hasn't changed (i.e. there is no "old" or "new").

Right. I was referring to the PM5 profiles which implement A2B/B2A mappings with the a and b elements mapped per the V2 spec and the L mapped to the V4 spec. The I1Profiler implements V4 printer profile A2B and B2A L,a,and b 16 bit encoding per the V4 spec. Thus PM5 is half baked though color engines apparently pick up the coding error and fix the L mappings.

V4 uses a mixed PCS encoding - a new one for some tags, the V2 one for other older tags. (it's a mess).

I agree that much of the Per I is a mess with reference mediums and such but the A2B1 and B2A1 tags are not part of that craziness.
At least I don't see any "mixing" in the PCS to Printer A2B1 and B2A1 specs. What are the issues with the tags? The one that directly affects the printer LUTs is the White Point tag but I don't see any changes there.