Luminous Landscape Forum
Raw & Post Processing, Printing => Colour Management => Topic started by: tvalleau on April 17, 2015, 12:48:21 pm
-
Hi folks
I come here once every few years it seems, and this time I'm trying to chase down a mystery with a group of profiles.
I recently paid a top-tier company to make custom profiles for me. I used PrintTool to print their target without Colorsync, Adobe or Epson messing with it. (The printer is a 9890.)
What happens with only these profiles, and it's across all of them, is this: 1) there is a color change in the cyan, and more obviously, 2) printed using Relative Colormetric (RC) intent, I get a banding of colors, some of which is quite obvious.
With perceptual, everything looks fine. Yes: I understand the difference between the two, and that's obviously in play here. The question is -why?-
I've switched between RC and percp tens of thousands of time over the years, using Epson profiles; my own profiles; paper mfgrs profiles, and I've never seen anything like this.
(My test image is the one with the clock in the upper left, and the sandbar in the lower right; strawberrys, yellow flowers, and the arches in the southwest. The pattern I see this in is at the bottom of the second column of images.)
I've uploaded a scan of the printouts, using the paid profile, an Epson profile and my own ColorMunki profile, all printed using RC intent. Epson and mine look as I've always seen; the paid one is out of whack.
I've dropped the profile into ColorThink, and it appears fine to me (although I could be missing something. I'm not a ColorThink expert.)
Now I've ordered an Xrite i1Photo Pro spectrophotometer to bump up my own quality level from the ColorMunki... and because I'm so puzzled by this, as is the provider of the paid profile. Someone in the QTR group on Yahoo suggested I come over here to see if anyone can explain why I'm seeing this only on these particular profiles, and have never seen it before.
I've attached the sample scan of the prints. It is also here: http://cl.ly/image/1o0e3O2r1t0L
Note the different color on column A & B; the "banding" in C but also in the magenta next to B and red next to D.
Here is the profile, if you'd like to see it: http://dlsi.biz/fc/Ep9890_EpPrmLustr.zip
If anyone here would care to take a moment to look at this, and enlighten me, I'd sure appreciate knowing what's going on.
Yosemite 10.10.3; 32 GB RAM; Photoshop CC (2014); PrintTool.
Thank you.
Tracy Valleau
www.valleau.gallery
-
Seems only two possibilities, either the target had some issues, or the profile creation had some issues. Have you spoken to the company that made the profile?
Curious why you used Print Tool instead of the Adobe Color Printer Utility (https://helpx.adobe.com/photoshop/kb/no-color-management-option-missing.html)? I would reprint the target from Print Tool and then from Adobe’s utility and see if visually there are any subtle differences. Also curious about the target, how many patches? Did you check the nozzles before printing the targets?
-
Thanks, Wayne. Greatly appreciate the reply.
Yes, I spoke with them at length, and they can't explain it.
I used PrintTool because the company's software generates a new target for each paper, with a special code on it. The default ppi embedded in the file is the wrong size to produce the needed 8x10 print, so each file needed to be resized (from 72 to 101 ppi.) I had 24 profiles made, and chose to use PrintTool because it allows for on-the-fly resizing, while CPU does not.
Yes, the heads are working just file, and yes, I did print out a chart using CPU, and eye-balled it in comparison to the PT version. They looked the same to me, although eye-balling probably isn't sufficient at this point.
There are 928 patches; 729 are color, and the rest gray-scale.
I can add this info:
their profiles default to RC, while most others I've seen default (in the header) to perceptual. (I realize that shouldn't make any difference, but then I shouldn't be seeing this anomaly either...)
Their target comes with an embedded sRGB indentification.*** That's likely because their instructions for printing without color manipulation on the Mac were out of date, using the old sRGB to sRGB technique that no longer works.
I'm fairly sure that didn't enter into the actual printing, since I printed one out using thru a paper profile, and it didn't look remotely the same., and, as I noted, the CPU and PT versions were (at least to my tired old eyes) identical.
*** this is likely incorrect. See below.
-
Their target comes with an embedded sRGB indentification. That's likely because their instructions for printing without color manipulation on the Mac were out of date, using the old sRGB to sRGB technique that no longer works.
You may want to try and map the target in ColorThink and compare against sRGB gamut. I suppose it’s possible that somehow in the process the files were actually clipped into the sRGB space.
-
Now that's an interesting possibility, Wayne. Thanks. I'll check into it, and report back. (I'm on my way out just now, so it may not be until tomorrow...)
Thanks.
-
Some thoughts…
By banding, I presume you are getting posterization, which is always a possibility with relative colorimetric.
928 patches isn’t a lot for a good profile, I’d look at it as a minimum. I do 1380 for even basic papers that I don’t use often.
Do the papers you have profiled have a high OBA content? That could make a significant difference, depending on the viewing conditions and the profiling process.
Brian A
-
Hello again...
Forgive me as a beginner at ColorThink, but here's what I did:
First, it appears that my comment about an embedded sRGB profile was wrong, and I simply must have gotten confused at some point, and accidentally applied one. My mistake.
Then, I opened the print target in PS: no embedded profile (sanity check)
Next, I open the print target in ColorThink : no embedded profile
In CT, I used colorSmarts Guid, chose by tool - grapher / compare image colors with device gamut
compare the EPL.tiff to sRGB profile
They are identical - a perfect match.
So, Wayne, you were certainly on the right track. I would not have expected the target to perfectly match the sRGB contour, but the (view by) points of the target lay perfectly on the (view by) wireframe of the sRGB system profile.
Now, in my ignorance, I'd have expected the target to be significantly larger than sRGB, else how would I ever get a profile that fully utilized the range of my printer (which is closer ARGB) otherwise?
Or do I simply not yet understand how a target is read and created? Shouldn't that target print (created by their software) been larger than sRGB?
Thanks for the hand-holding. it's humbling to think that I clearly understand this stuff, and find myself quite so lost. I appreciate the help.
-
Next, I open the print target in ColorThink : no embedded profile
Then it’s impossible to say it’s in sRGB or otherwise to make any additional comparisons. CT doesn’t know what the RGB values are without an embedded or assigned profile, how could it? It can’t plot or provide Lab values either. IF you assign sRGB to the numbers, they will match, well sRGB.
I would not have expected the target to perfectly match the sRGB contour, but the (view by) points of the target lay perfectly on the (view by) wireframe of the sRGB system profile.
Assign Adobe RGB (1998) and plot it against that profile. You’ll get the same results if I follow what you’re doing.
CT isn’t going to help you a lick unless you have the measurement data used to build the profile.
-
Hi Brian, and thanks for jumping in.
Yes, I considered calling it posterization at first, as the vivid light magenta column certainly looks like that. But closer examination of the image printed thru the resulting profile shows (in the red and green) areas where what should have been a smooth transition from light to dark, instead move from light to dark, to light, to dark, to light again, as if the gradation was re-arranged in small chunks.
True, that could be a form of posterization fooling my eyes, so I won't seriously disagree with the use of that term.
I've seen a lot of posterization in my time, and that said, what I have not ever seen is this happen like this from a RC conversion. As I understand it, colors within the gamut are not moved at all, and only colors outside the gamut are moved to the closest neighbor within the gamut. (To the edge, if you will). So I'd have expected -most- of the color to be within gamut, and no posterizing for the vast majority of the tone range.
That has been my experience with all the other profiles I have: I can switch between perceptual and RC with relative impunity, and certainly without what I'm seeing in the image I posted here.
-
Thank you, Andrew. That completely makes sense. I might be able to come closer on Monday, when the new i1 Photo Pro spectrophotometer gets here. I'll see if it's possible for me to make a profile from one of the printed targets, and if so, how it compares in the same image test.
Do you have any suggestions as to how else I might nail down whether it's the printed target, or the profile that was generated from it?
I can use the paid profiles with perceptual intent, of course, but some of my work requires RC...
-
snip
928 patches isn’t a lot for a good profile, I’d look at it as a minimum. I do 1380 for even basic papers that I don’t use often.
Do the papers you have profiled have a high OBA content? That could make a significant difference, depending on the viewing conditions and the profiling process.
Brian A
I neglected to answer the rest of your questions. Yes: some of the papers have significant OBAs, some have none; however the effect, to greater to lesser degree, happens with all profiles on all the papers I've retested so far. The example I posted has what I'd call "medium" OBAs - it's Epson Premium Luster.
Of course, I'm using their software to generate a target for their automated reader, so I have no choice in the number of patches. Come Monday, I'll remember your advice as to the 1380 minimum. Thanks for that.
-
One last thing I tend to forget, is that not only does that posted image show "banding/posterization" but the cyans are significantly darker than are those printed thru the Epson and ColorMunki profiles. I'd assume that the two symptoms are related...
-
Just to make sure then: ColorThink will apply an sRGB profile to an image file if that image file is untagged? I ask because when I open the (apparently) untagged target and compare it to the AdobeRGB space, it (the target) is showing the smaller sRGB graph points within the larger ARGB...
Also, I've removed the Adobe CPU vs PrintTool from the equation by printing the target with each; scanning the print and layering the results in Photoshop, using difference. Got a nice, pure black result, so the printing software is generating the same thing regardless of which is used.
-
Just to make sure then: ColorThink will apply an sRGB profile to an image file if that image file is untagged?
It can’t produce Lab values which is the key. It ‘sees’ the RGB device values (R23/G66/B90) but those values are meaningless without a profile.
Take a target, assign two different profiles to the same numbers, ask to Plot List. You’ll get a different plot from identical device values but of course, differing Lab values.
Take the target and plot with and without sRGB, it appears the two plot the same so yes, it appears CT assumes sRGB for untagged numbers.
-
Thanks again, Andrew. CT's automatic use of sRGB on an untagged file is what had me confused before. The rest of your explanations jibe with what I (thought) I knew before, so that's of some comfort... ;)
At this point, with the "banding" and the color shifts, I'll probably be better off waiting for the new i1Photo to come on Monday and see if the same thing happens to me.
Meanwhile, I'm happy to pursue other suggestions.
(My point in pursuing all this is two-fold. If it's somehow my error, I'd naturally want to know that, and 2) if their equipment is out of calibration or suffering the effects of age, I'm sure they'd want to know that as well.)
Thanks again everyone.
-
Maybe it's me, but your posted image doesn't show any problems, and the Paid looks better than the Epson.
-
Maybe it's me, but your posted image doesn't show any problems, and the Paid looks better than the Epson.
Really ? If you can't see that step in the magenta gradient marked as C you shouldn't comment any further on this.
-
There are 928 patches; 729 are color, and the rest gray-scale.
This fact makes me suspect quite strongly that they are using the SpyderPrint system. I own it and have abandoned it, and I don't think a professional service should be using it.
My thoughts on the system are described here:
http://forum.luminous-landscape.com/index.php?topic=78431.msg630093#msg630093
-
Nope: not SpyderPrint (which I tried many years ago and abandoned instantly as well.)
As I noted, the profiles in question were made by a very well respected company, whose name would be familiar to everyone here, and I can pretty much assure you are not using SpyderPrint. I have not mentioned the name because I'm still assuming it's something on my side of the equation.
I expect my new hardware to arrive in a few hours, and I'll try my own profile on Epson Premium Luster and see if the issue shows up in that one. (Of course, that won't prove anything about why theirs exhibits the problems.... )
-
For those who have been using spectrophotometers for a while now: is it possible for the lamp to age, and induce a color-shift during reading? Could there be any other kind of issue with their hardware that might cause this?
-
For those who have been using spectrophotometers for a while now: is it possible for the lamp to age, and induce a color-shift during reading? Could there be any other kind of issue with their hardware that might cause this?
Lamps do age, but that's what the calibration tile is for and why you calibrate it before you use it. The long term accuracy largely depends on the stability of the reference tile, and the stability of the optics with regards wavelength (which is why the i1pro2 does a wavelength calibration as well as white tile measurement).
I have seen some other effects in older lamps such as a "memory" or "warming" effect where the recent on/off timing affects the current reading more than usual. I'm not sure if that's something peculiar to the particular (older) instrument I was looking at, or a consequence of lamp aging though.
-
Thanks. I appreciate that bit of knowledge.
-
non-update: I said I'd have profiled the Epson paper by today using my new i1Pro2 today, but I lied. The device didn't arrive until this morning, and I have a meeting tonight. Mañana...
-
Well, I broke out the new i1Pro2 and stumbled thru my first printer scan. (I say stumbled because a) the instructional videos are 4 years out of date; 2) starting in "basic" mode seemed appropriate for a newbie to this particular hardware, but I went down the rabbit hole by trying to adjust the patch sizes (after getting a warning their own (!) default sizes were too small.
Finally, I gave up and just did a MO default all the way across the board, (800 patches) and it scanned perfectly. (In fact, having done this kind of thing with different hardware years ago, I'd have to say I was impressed at how well it performed: I didn't mis-scan a single line.)
OK: so much for the back story. It's time to cut to the chase...
... I printed the same image on the same paper using my newly created profile, and relative colormetric rendering intent... and it looks perfect. The cyans are the correct tone and the "banding" (the "C artifact") is completely gone.
Dropping my new one, and the old paid one into CT, I can see that mine is larger, particularly in the green/blues and generally overall.
So, now I have Epson's profile; my own ColorMunki profile (sold the CM yesterday, now that I have an "oooohhhh shiney" serious tool) and today's newly created profile. None of these exhibit the original issues with RC intent. Only the paid one is wrong.
Everyone here would recognize who did the paid profile for me (and some of you may have easily found it out.)
They are fine folks, and I'd like to pass this along to them, but at this point, the only conclusion I can come to is that there is something wrong with their profiles (since every one exhibited the same RC issue, to a greater or lesser extent.)
Does anyone here have any further comments, before I head off to recreate my profiles (for an exhausting -third- time now) ?
Thanks to everyone for your interest and advice.
OH... and if anyone has tips I might use for my new i1Pro profiling (I'm hand pulling the scanner; just can't afford the automated toy right now) I'd love to hear them as well.
Once again: thanks.
Tracy
www.valleau.gallery
-
With the old i1Pro, hand scanning was always a problem for me but something has changed a lot. I rarely get an error when scanning a target now. The old i1 was so bad that I bought the io table. It was that or give up profiling. The new device also seems to give more consistent results.
I'm sure that wiser heads will comment but 800 patches sounds way too few. I scan 2000 patches which is 3 A4 pages. My understanding is that around 2000 is where more doesn't help. I also save my scanned data as PM5 txt which makes it easy to drop into CT.
-
I'm sure that wiser heads will comment but 800 patches sounds way too few.
Even the old TC918 target can build a very good profile. Heck, with as few as 100 (50/50), ColorMunki is darn impressive. Sure, use more, but that doesn't mean, depending on the behavior of the device, 800 patches will suck.
-
Thanks, Jeff. As I explained, this was my first hand scan with the new equipment, so I used the most basic setting the (new) software supplies. That, and I wouldn't think that more samples would cause the negative effect I've reported.
Yes, that said, when I'm making "real" profiles for daily use, I'll use more patches. I just wanted to quickly see if the problem remained.
It did not.
-
Even the old TC918 target can build a very good profile. Heck, with as few as 100 (50/50), ColorMunki is darn impressive. Sure, use more, but that doesn't mean, depending on the behavior of the device, 800 patches will suck.
I thought you would answer, Andrew. My understanding is that 1000 to 2000 patches is a good number but I got that a while back. Does that still hold? I understand that a good profile can be had with less but, at what point does more stop improving the result?
-
I thought you would answer, Andrew. My understanding is that 1000 to 2000 patches is a good number but I got that a while back. Does that still hold? I understand that a good profile can be had with less but, at what point does more stop improving the result?
It depends, sorry. But the point is, 918 can work. I'd probably agree, something around 1000-2000 is a sweet spot. Depends on the software building the patches too, they are not all equal. There's a point of diminishing return. So let's say, 1500 or so.
-
To "choice of software" I'm using i1Profiler 1.6.1 (XRD version 2.3.8.19)
...if that matters...
-
It depends, sorry. But the point is, 918 can work. I'd probably agree, something around 1000-2000 is a sweet spot. Depends on the software building the patches too, they are not all equal. There's a point of diminishing return. So let's say, 1500 or so.
Thanks Andrew. I just went hunting around and found the Bill Atkinson profiles that were so popular a while back. Are they still a good choice or are the i1Profiler generated charts a better match these days. My memory is that they were highly regarded at the time.
-
...
OH... and if anyone has tips I might use for my new i1Pro profiling (I'm hand pulling the scanner; just can't afford the automated toy right now) I'd love to hear them as well.
I have found, at least on fine art papers, two sheets of full patches, and one sheet of optimization patches, produces more accurate profiles than four sheets of original patches. There is no increase in gamut, but better linearity.
Brian A
-
Excellent, Brian. Exactly the kind of useful tip I was looking for.
(Having heard no more here, I'll be pointing the folks who did my paid-for profiles as this discussion... and asking for a refund.)
I'm sorry that there was no gain in knowledge (for the group) to be had from this little adventure, but my thanks again for the courtesy from everyone here.
I'm off to see the Wizard ;)
-
Well, it's been a while, and I thought it was time I round out this topic, since I started it. A few weeks ago, the company sent along a second profile for me to check, and the banding simply changed places, and (interestingly) moved to a new color. I offered to continue helping track down the problem, but they "went silent" - so I suppose they think it was something applying only to me, or just plain lost interest. Maybe it was me; maybe the other who may use their service simply don't have, or never saw, the issues I reported with RC. The company did, at my request, refund my profile subscription fee.
I'm perfectly happy with the i1 Photo Pro, scanning 1600-patches for each profile, and the results only make me wish I'd done this years ago.
Thanks once again for the interest, help and courtesy. Best wishes to one and all.
-
With the old i1Pro, hand scanning was always a problem for me but something has changed a lot. I rarely get an error when scanning a target now.
The i1pro2 has a tracking ruler (technology borrowed from the DTP20 presumably), primarily to match the two passes when you do a UV scan (to compute M0, M1 etc.), so if they use that for normal scanning as well, the patch recognition should be more robust.
-
I believe the company is using an i1iSis to generate the profiles (from an 800-patch chart.)
I, on the other hand, print out the 1600 patch and using the sliding ruler setup you mention, can turn out a profile in about 10 minutes.
-
The i1pro2 has a tracking ruler (technology borrowed from the DTP20 presumably), primarily to match the two passes when you do a UV scan (to compute M0, M1 etc.), so if they use that for normal scanning as well, the patch recognition should be more robust.
Thanks Graham. It is a vast improvement of the original.
-
FWIW, I did notice a significant increase in i1Photo Pro reliability (of the passes and calibration) when I replaced the bundled USB cable with a better one. YMMV.