No color management

[Stephen Ray]

My main point is that there is no special reason for manufacturers to all use the same L*  RGB targets and they clearly do not.

Thanks for your observations. It would be interesting to learn what others are measuring from the same machine models using the same media.

[BradSmith]

Think of it this way:  You are sighting in a rifle - you take 5 shots at a target and they are all very tightly clustered (precise) but high and left of the target's bull's eye.  So the rifle is precise (tight cluster of hits) but not accurate (the bull's eye is the reference).  So you adjust the scope via an ICC profile and, when done correctly, the tight cluster of shots is still tightly clustered (precise) but now it is tightly clustered on the bull's eye (accurate).

In this analogy, you do not adjust the rifle (the printer) you adjust the scope (the ICC profile) to optimize the inherent precision of the rifle so that it is both precise AND accurate.

kirk

This seems to me to be an excellent analogy of what is going on.  Well done Kirk

[Doug Gray]

The other way around.  The printer manufacturer makes sure that if you feed it an RGB triplet it will print that triplet in a repeatable, well characterized manner so that if you feed it the triplet 100 times, the variation between prints will be small, within some quantifiable tolerance for coverage uniformity, measured reflectance or density, etc.  That is precision, regardless of whether or not that printed value represents what the user desires (the "actual" value).

Accuracy is the printer's ability to print an input that represents a reference value - if you feed the printer an RGB triplet that represents L*50, it should print at L*50.  If it does, it is accurate, if it does not it is not accurate.  Color management and ICC profiles provide the method by which the printer's native color (which is precise and repeatable) gets transformed into an accurate representation of a specific target color - that is, the difference between the print and the reference (usually expressed as deltaE) is within some tolerance.  If the difference between the print and the reference falls within the tolerance, the printer is "accurate."

Think of it this way:  You are sighting in a rifle - you take 5 shots at a target and they are all very tightly clustered (precise) but high and left of the target's bull's eye.  So the rifle is precise (tight cluster of hits) but not accurate (the bull's eye is the reference).  So you adjust the scope via an ICC profile and, when done correctly, the tight cluster of shots is still tightly clustered (precise) but now it is tightly clustered on the bull's eye (accurate).

In this analogy, you do not adjust the rifle (the printer) you adjust the scope (the ICC profile) to optimize the inherent precision of the rifle so that it is both precise AND accurate.

kirk

Kirk,
It is the location of the bullseye that varies with printers/papers. My Epson 9800, unlike the Canon 9500 has always printed consistently and it's profiles are quite accurate except at lower luminances. One can reverse the canned profiles of any printer to determine what the printer was designed to print when given an RGB triplet. For instance the 9800 reports it will print L*=32 for RGB 128,128,128 and L*=69 for RGB 200,200,200.  It actually prints L*=34 and L*=69. In other words the 9800 is printing very close to it's design target. The reason the Epson prints L*34 when the canned profile reports 32 is that they incorporate BPC in their A2B1 and B2A1 tables. This is not in accord with ICC specifications but a common practice none the less. The result is reported lower L* than is actually printed at low L* values. It gets much worse lower. It reports L*=0 for RGB 0,0,0 when it should be reporting L*=4, its actual black point.

[Stephen Ray]

Is there a benefit of not targeting a logical bullseye of L*50?

[Doug Gray]

Is there a benefit of not targeting a logical bullseye of L*50?

I suspect there is. The darker colors are a complex mix of overlapping inks and there could well be more non-linearity in these regions. Lighter colors are more likely to behave more smoothly since there are fewer dot on top of dots as well as more uninked areas. By setting the RGB 128,128,128 at lower L* they leave more room in the device RGB space for dealing with these issues.  But it's just a guess. They clearly are doing it on purpose so it has to provide some benefit.

For what it's worth, the Canon 9500II canned profiles are not nearly as accurate as the Epson and the actual printer colors for RGB 128,128,128 are a fair amount lower than the canned profiles indicate which is over 50. Another factor that might be at play is the black point of the 9500 is quite high, close to L*=8 on their glossy media.

[kirkt]

The bull's eye/rifle analogy was to describe precision versus accuracy, not something specific about a particular printer.  Hopefully it was helpful.  Precision and accuracy get used interchangeably in colloquial use, but they have specific definitions in engineering and scientific context (i'm an engineer). 

kirk

[Doug Gray]

The bull's eye/rifle analogy was to describe precision versus accuracy, not something specific about a particular printer.  Hopefully it was helpful.  Precision and accuracy get used interchangeably in colloquial use, but they have specific definitions in engineering and scientific context (i'm an engineer). 

kirk

kirk,

Your bullseye analogy is quite good but invoked some additional discussion as to the fact that each printer maker has, for whatever reasons, targeted their unmanaged color RGB space differently so the RGB 128,128,128 does not produce L*=50. People naturally think of a bullseye as centered and hence should target L*=50. whereas each printer design may and do have different mappings. However, it's also obvious from your comments you know this and were not implying that RGB 128,128,128 should produce L*=50. I found your comments quite clear and well stated.


As an aside, here's one way to identify how a manufacturer maps their RGB device space if, for instance, we wish to know what color a printer is designed to print 128,128,128 as.

1. Create an image patch in Photoshop filled with RGB 128,128,128.
2. Assign the printer/paper profile to that image.
3. Convert the image to Lab colorspace using Abs. Col. Intent.
4. Examine the Lab value. This is what the printer manufacturer has targeted.

To the extent wear and tear, manufacturing variations, and such come into play the printer actual color printed may vary from this. Custom profiles provide a way to compensate for these.

[Stephen Ray]

What I am finding from cursory checks around the inter webs is L* targets for middle gray, either 128 or 50% dot is from ~ L*50 to L*60. Some sources is a research paper from CalPoly about G7, various RIP targets, and performing Doug’s method of tagging printer profiles then converting them to LAB using Abs Col Intent. The profiles I tested are Canon IPF1000 provided for popular 3rd party papers. Apparently the IPF1000 has an on-board calibration process. Maybe most pro printers from Canon have this now. I don’t know.

I’ve just spent exactly 1 minute at Canon’s “Experience Center” showroom, repair and customer service facility but no one was available to answer a technical question at the technical center. So much for that, however I will return soon just because it’s walking distance. My “experience” was short of what my expectations would have been years ago but nowadays I expect my experience to be exactly what it was this morning. Quote, “Please check our website. Please come again.” Epson announced last year they want to build a showroom. May be my next stop.

[Doug Gray]

Good idea to keep in mind that L*=50 is only 18% reflectance or about 20% of what is reflected off unprinted paper.

No idea why some manufacturers chose higher or lower points. I've looked at a few Canon's and they are all just over 50. Epsons tend to be under 40. I think it just depends on how the designers decide to do the rasterization.

An important question outside the neutral colors: What is the color difference distribution for single bit changes across the device RGB space. At least for 8 bit drivers. And how does the L* target influence this distribution?

Curious as to what info and rationale you get from your inquiries.

[nirpat89]

As an aside, here's one way to identify how a manufacturer maps their RGB device space if, for instance, we wish to know what color a printer is designed to print 128,128,128 as.

1. Create an image patch in Photoshop filled with RGB 128,128,128.
2. Assign the printer/paper profile to that image.
3. Convert the image to Lab colorspace using Abs. Col. Intent.
4. Examine the Lab value. This is what the printer manufacturer has targeted.

To the extent wear and tear, manufacturing variations, and such come into play the printer actual color printed may vary from this. Custom profiles provide a way to compensate for these.

Doug: 

I thought I will try this sequence and see what happens.  It would be interesting to do this on various printers and papers and compare.

1 and 2, I understand.  In 3, for converting to Lab, Photoshop simply converts to Lab (image>mode>check Lab.  There is no option for intent.  Is there an another way?

Anyway, if I simply have a block of 128-128-128, assign it a printer profile and change the mode, the results are as attached.
The end result is a Lab of 44, -3, -2.

Does that sound right?  Wouldn't this number be representative of what Photoshop sends to the printer and not what the printer print it as.  I am not understanding it.

I used the profile provided by the paper manufacturer (Canson Rag Photographique on Epson SC P400.) 

:Niranjan.

[Doug Gray]

1 and 2, I understand.  In 3, for converting to Lab, Photoshop simply converts to Lab (image>mode>check Lab.  There is no option for intent.  Is there an another way?

Yes, Edit->ConvertProfile, This will give you conversion options. The Image->mode>Lab does not use Absolute Intent and has no conversion options.

[nirpat89]

Yes, Edit->ConvertProfile, This will give you conversion options. The Image->mode>Lab does not use Absolute Intent and has no conversion options.

Got it.  Did it that way.  Now the block is (46, -3, -1).