Spyder 5 print Lab values

[Rhinetone]

Hello folks,

I have a question regarding printer profiling with a Spyder 5 Print.

At the end of the measurement process, I can export a text file with all the measured Lab values. My problem is, I have nothing to compare them against. I understand that there is no 'reference' value, because the system measures the actual print.
But I can't easily verify if the measurements are correct.
I don't find the suggested way of comparing the measured value to the RGB patch on screen in split mode very accurate. Yes, I can detect crude errors, but I still find it a bit tedious and not very precise.

I want to find a way to quickly zoom in on measurement errors. Here's my thinking: If I do two measurements, export the Lab values to a text file and then import the data into an Excel spreadsheet, I can easily create a formula to find the measurements where the delta values are outside of the (yet to be determined) tolerance.

So how should I define these delta values? To give you an example:
My first measurement of the first patch was L=16.79 , a=0.59 and b=4.32
The second measurement was L=17.04 , a=-0.3 and b=5.54

So if I set the calculations in the spreadsheet to flag me for all the individual patches, where the difference from one measurement the the other was more than 0.5, for example, in either luminance or the a or b channel and I could just quickly re-measure those patches. So the example above would be a 'fail', because both the a and b channel are outside of this tolerance.
The measurements of another random patch with values of L=39.11, a=28.23, b=-44.34 and L=39.04, a=27.82, b=-44.11 are a 'pass', because all values are within the set tolerance of <0.5.
 
First question: Would that work or am I forgetting something?
Secondly: What difference between two measurements would you find acceptable in each channel?
How should I set my tolerance? I have no idea if the values above are typical or not. I personally feel that the difference between the two measurements in the first example is a bit high, but I have very limited experience with Lab values.
I couldn't find anything in the Spyder 5 documentation about the accuracy of the device.

Any opinion on the above would be much appreciated.

Kind regards

Andy

[Mark D Segal]

Hello folks,

I have a question regarding printer profiling with a Spyder 5 Print.

At the end of the measurement process, I can export a text file with all the measured Lab values. My problem is, I have nothing to compare them against. I understand that there is no 'reference' value, because the system measures the actual print.
But I can't easily verify if the measurements are correct.
I don't find the suggested way of comparing the measured value to the RGB patch on screen in split mode very accurate. Yes, I can detect crude errors, but I still find it a bit tedious and not very precise.

I want to find a way to quickly zoom in on measurement errors. Here's my thinking: If I do two measurements, export the Lab values to a text file and then import the data into an Excel spreadsheet, I can easily create a formula to find the measurements where the delta values are outside of the (yet to be determined) tolerance.

So how should I define these delta values? To give you an example:
My first measurement of the first patch was L=16.79 , a=0.59 and b=4.32
The second measurement was L=17.04 , a=-0.3 and b=5.54

So if I set the calculations in the spreadsheet to flag me for all the individual patches, where the difference from one measurement the the other was more than 0.5, for example, in either luminance or the a or b channel and I could just quickly re-measure those patches. So the example above would be a 'fail', because both the a and b channel are outside of this tolerance.
The measurements of another random patch with values of L=39.11, a=28.23, b=-44.34 and L=39.04, a=27.82, b=-44.11 are a 'pass', because all values are within the set tolerance of <0.5.
 
First question: Would that work or am I forgetting something?
Secondly: What difference between two measurements would you find acceptable in each channel?
How should I set my tolerance? I have no idea if the values above are typical or not. I personally feel that the difference between the two measurements in the first example is a bit high, but I have very limited experience with Lab values.
I couldn't find anything in the Spyder 5 documentation about the accuracy of the device.

Any opinion on the above would be much appreciated.

Kind regards

Andy

Hi Andy,

It isn't clear to me what you are really trying to do. Are you (a) trying to ascertain the accuracy of how the printer prints colours using the profile you created, or (b) are you trying to see whether the profiling patches you just measured for creating the profile have the same values as in their reference file?

If you are trying to do (b), don't bother because it isn't a useful exercise. The measured values of the profiling patches will rarely be the same as their file reference values because the printer does not print them accurately without a profile. The whole purpose of the profile is to characterize the printer's reproduction of those patches absent any colour management for the purpose of creating a profile that assists with the colour reproduction correction processes applied in the print pipeline. If you are trying to do (a), you need to print a set of colours with known reference values using Absolute Rendering Intent, then measure them from the print with your spectrophotometer and identify the differences. Small differences are not important, larger ones may be. You will get different differences between file values and read values depending on which definition of dE you use, as say between dE(76) or dE(2000). Using the latter, differences of less than 1.0 are supposed to be imperceptible to humans.

[Rhinetone]

It isn't clear to me what you are really trying to do. Are you (a) trying to ascertain the accuracy of how the printer prints colours using the profile you created, or (b) are you trying to see whether the profiling patches you just measured for creating the profile have the same values as in their reference file?

Hi Mark,

Many thanks for your reply.
Basically, I'm looking for a quick way to catch any measurement errors whilst creating the profile.
I don't find it easy to check for errors on screen. The Spyder software provides a function where you can compare the measured value to the RGB patch on screen, but it's not easy to spot if one measurement is slightly off.

That's why I'd rather work with the Lab values. I understand that it's not much use to compare my measurement with a reference value, because - as you said - at this point I don't have a profile yet, so I need to work with what the printer gives me.
Because of that, I thought if I create two measurement files, I can compare the two and quickly zoom in on any patch where there is a large difference between the first and second measurement. This would indicate to me that one of the measurements is not accurate.
Taking a third measurement of this particular patch would tell me (more or less), which one was accurate.

I'm just not sure how much tolerance I should allow when comparing two values. For instance, if the L in one measurement is 20 and the second time round it's 50, I am pretty sure this would indicate a measurement error. I would not necessarily know which one is correct, but if I measure it a third time and it reads 22, I can be pretty certain that the first measurement is correct.

I have also compared two measurement files using dE(76). The math involved is a bit easier to understand than with alternative colour difference calculations. I think a value of <2.3 is thought to be acceptable.
Comparing my two measurement files of 729 patches, there were just over 60 patches where the dE(76) was larger than 2.0. That's quite a lot, I think, but it may be perfectly normal - I don't know. But this would mean that over 8% of all patches show a difference between measurements which would (just) be visible. I know this is all a bit theoretical, and ultimately I just want to get the best possible printer profile. But from what I have discovered so far, I don't think that measuring just once will give me that.

Kind regards

Andy

[Mark D Segal]

Hi Mark,

Many thanks for your reply.
Basically, I'm looking for a quick way to catch any measurement errors whilst creating the profile.
I don't find it easy to check for errors on screen. The Spyder software provides a function where you can compare the measured value to the RGB patch on screen, but it's not easy to spot if one measurement is slightly off.

That's why I'd rather work with the Lab values. I understand that it's not much use to compare my measurement with a reference value, because - as you said - at this point I don't have a profile yet, so I need to work with what the printer gives me.
Because of that, I thought if I create two measurement files, I can compare the two and quickly zoom in on any patch where there is a large difference between the first and second measurement. This would indicate to me that one of the measurements is not accurate.
Taking a third measurement of this particular patch would tell me (more or less), which one was accurate.

I'm just not sure how much tolerance I should allow when comparing two values. For instance, if the L in one measurement is 20 and the second time round it's 50, I am pretty sure this would indicate a measurement error. I would not necessarily know which one is correct, but if I measure it a third time and it reads 22, I can be pretty certain that the first measurement is correct.

I have also compared two measurement files using dE(76). The math involved is a bit easier to understand than with alternative colour difference calculations. I think a value of <2.3 is thought to be acceptable.
Comparing my two measurement files of 729 patches, there were just over 60 patches where the dE(76) was larger than 2.0. That's quite a lot, I think, but it may be perfectly normal - I don't know. But this would mean that over 8% of all patches show a difference between measurements which would (just) be visible. I know this is all a bit theoretical, and ultimately I just want to get the best possible printer profile. But from what I have discovered so far, I don't think that measuring just once will give me that.

Kind regards

Andy

When you characterize the printer, which is the process of printing the profiling target and measuring it, the measurements you take of that printed target indicate what the printer produced in response to being fed the patches. It is always inaccurate for two reasons: (1) printers do not reproduce colours "accurately" (i.e. the same as the target reference values) without a profile, and (2) spectrophotometers - at least the kind we normally use - almost never return the identical value for successive measurements on the same spot of the same patch. There are always small differences. Some profiling packages allow the calculation of averages of a number of readings to handle these discrepancies. What I am suggesting to you is not to bother about such accuracy AT THIS STAGE (i.e. when creating the profile). Take one set of measurements, or averaged measurements of the printed profiling target if your software allows, make your profile from those measurements, then test your profile by printing a target of known values to see whether it is allowing you to print accurately enough for your needs.

[digitaldog]

Hello folks,

I have a question regarding printer profiling with a Spyder 5 Print.

At the end of the measurement process, I can export a text file with all the measured Lab values. My problem is, I have nothing to compare them against. I understand that there is no 'reference' value, because the system measures the actual print.

There would be with other systems but perhaps not the Spyder. For example, with X-rite products, there's a reference file that is used to build the targets you printed. That would provide the reference Lab values. It's actually not all that useful depending on what you're trying to actually compare. Sure, reference vs. measured is one metric but what you really want in most cases is to compare the measured color with the predicted color from the profile or the round trip as discussed below, using a technique defined by Bruce Fraser, the idea outlined in a PDF I produced awhile ago:


http://digitaldog.net/files/How%20to%20compare%20the%20deltaE%20of%20ColorLists.pdf

[Doug Gray]

Andrew and Andy,

Patchtool is also a good product and can easily create (and provide metrics on the result) a set of patches across a printer's gamut with a desired stepping in L*, a* and b*.  Quite handy to test round tripping with.

Andrew: nice writeup for peeps interested in an in depth understanding and measurement of their printer profiles.

[GWGill]

But I can't easily verify if the measurements are correct.

If by "correct" you mean "not a misread", then repeating the set of measurements in a different order, and then checking the delta E's between the data sets is one way of doing this.

Some profiling packages may give you an indication of the self consistency/smoothness/self fit of the patch set when you create the profile, and excessive self fit errors can be a hint that there is a misread in the patch set.

Of course checking the absolute accuracy of the measurements requires a reference instrument to compare against.

So how should I define these delta values? To give you an example:
My first measurement of the first patch was L=16.79 , a=0.59 and b=4.32
The second measurement was L=17.04 , a=-0.3 and b=5.54

Standard way is to compute the total delta E, which is simply the euclidean distance between the two points in Lab space.

[Rhinetone]

Hello again,

First of all, many thanks for all your contributions.
Phew, what can I say - I am learning a lot!

If by "correct" you mean "not a misread", then repeating the set of measurements in a different order, and then checking the delta E's between the data sets is one way of doing this.

Yes, this was exactly my primary aim.

Patchtool is also a good product and can easily create (and provide metrics on the result) a set of patches across a printer's gamut with a desired stepping in L*, a* and b*.  Quite handy to test round tripping with.

Could you explain the term 'round tripping' to me in this context?

Sure, reference vs. measured is one metric but what you really want in most cases is to compare the measured color with the predicted color from the profile or the round trip as discussed below, using a technique defined by Bruce Fraser, the idea outlined in a PDF I produced awhile ago.

Many thanks for providing the link. I've not had time yet to really study the document, I will need a bit more time to try and digest it.

... then test your profile by printing a target of known values to see whether it is allowing you to print accurately enough for your needs.

By 'known values', do you mean print it one of the recognised printer test files? Or is there a file available you can recommend. Maybe there isn't, because if I understood Andrew's document correctly, the first steps are all about creating the 'known values'.

Well, I have now completed the profile. And with a bit of fun with dE calculations and euclidean distances, I am pretty sure, that the profile does not contain any measurement errors.
So far so good. Next, I've printed a test image - I'm sure you're all familiar with it. It's the one with the boat, the fish, the coloured fabric and Bill Atkinson's balls.

There are issues. The colour is pretty good, apart from Cyan. There is quite heavy banding with all colours, there seems to an abrupt switch from 'dark' to 'light'. That's maybe not surprising since my printer doesn't have any Light Black inks.
Btw, my printer is an Epson SC-P 400 and the paper I am profiling is an Innova 315gms Soft Textured Bright White Cotton paper.
I have attached a picture of these balls. Now I know that it's rare to get a perfect print of these balls, since it's so taxing. And looking at the pictures included with the test file, I am pretty happy with the result.

But my question is:
Looking at the attached picture, where do you see the opportunity to optimise the profile? As I said, I'm well aware that my printer may have certain limitations I won't be able to overcome. I struggle to interpret the resulting print a little - how good or bad is what I am seeing?
Secondly, what would be the suggested workflow to improve or optimise an existing profile? I read about linearisation for b&w profiles, but what would you do for a colour profile? Is there a way to improve my existing profile?

Kind regards

Andy

[Mark D Segal]

Andy, a target with known values simply means a set of patches whose colour values you know both from reading them in Photoshop from the info palette and by making sure those are the same numbers in the reference file for that target.

Another critically important point for testing the accuracy of the printer/paper/profile combination is that all the target colours should be within the gamut of the paper/printer. If they are not, you will be measuring inaccuracy that comes from how the Rendering Intent remaps out of gamut colour to fit at or within the gamut boundary. That's fine if you want to evaluate how the profile performs with out of gamut colours, but it's not fine if all you are looking for is the accuracy of colour rendition for those colours that the printer/paper should be capable of reproducing without remapping from an out of gamut space to an in gamut space.

[Doug Gray]

Could you explain the term 'round tripping' to me in this context?

The most basic question applying to printer profiles is how accurately is the image color printed? These profiles have two tables that map colors. The AtoB1 table makes an estimate of what color a specific RGB value at the device prints. The BtoA1 table, OTOH, is the table that determines what RGB value is sent to the printer from an image value.

When a profile is made by measuring a patch set of RGB values most tools will provide various statistical info on how accurate things are by comparing the known RGB values used to create the profile image with the LAB values the profile produces using the AtoB1 tables. Since these tables are directly generated from the printed patches used to create the profile they are typically quite accurate. But what concerns the printer is how accurately their image colors print and those are determined by the other table, the BtoA1 table.

Secondly, to determine printer accuracy one should use a completely independent set of colors  and not the ones used to create the profile.

Patchtool lets one generate a set of LAB colors in even steps, say (40,30,-20; 40,30,-10; 40,30,0) etc. that is inside the printer's gamut. These can be used to create an image which is then printed and scanned with a spectro to determine the accuracy of the BtoA1 tables. The statistics from this is effectively testing the profile generation algorithms in a manner of concern to the printer: "How accurate are my prints."  It's effectively the result of using one patch set to generate the profile using printer device RGBs then using that profile to print known colors which is from the direction of image colorspace, which is converted w/o loss to LAB, and then sent to the printer. The resulting variation, using two different data sets scanned with a spectro, is a kind of round tripping.

Another kind is using the in-gamut patch set image, converting in Photoshop to device space, then converting back to the original image space and comparing the differences statistically. This roundtrips the profile (image->BtoA1->AtoB1>image2). The statistical differences between the two is an indication of the profile quality.

Both of these are made easier with Patchtool's statistical and patch generation capabilities.

[Rhinetone]

Just wanted to thank all you guys who have contributed to this thread.

Your input was invaluable.
I initially posted this because I wasn't sure about the quality of the profiles I was creating.
It wasn't much more than a gut feeling at first. The process of profiling worked as you would expect, but I just felt that the resulting profiles were not as good as they could (should) be.

Not having a lot of experience with printer profiling, I needed to generate some data to substantiate this feeling. In this process, all your contributions were most helpful.
Many thanks for sharing your knowledge and expertise.

With the data I have gathered, I contacted Datacolor. After a few email exchanges and further tests, they have now reached the conclusion that my sensor is faulty and I am awaiting a replacement.
Just wanted to mention that the Datacolor tech support was very responsive and helpful. I am even getting a free replacement although technically the sensor is just out of warranty. Top marks!

Andy