QTR + MIS Carbon inks: Banding-Free Gradients ?

[Ken Lee]

I've been trying to print some photos which contain large smooth gradients in out-of-focus areas.

Using the available QTR tools to make profiles (including linearizing tools provided by Richard Boutwell (http://www.bwmastery.com/)) I have been unable to make prints whose gradients are truly smooth. The banding is not blatant, but present enough to spoil the image, especially on large prints.

I suspect this is because measuring only 21 steps is an inadequate approach. Also, my consumer-grade Spyder spectrophotometer gives me imprecise and varying readings: I have to average them out.

Prints of ordinary subjects look perfect. This problem only appears in images which contain substantial gradients. This isn't about print head alignment or cleaning cycles, this is about the quality of the profile.

Somewhere I read that Piezography profiles are custom-linearized to exceed the imprecision inherent with ordinary QTR tools. This might explain why they are available for only a limited selection of papers.

Has anyone printed out any gradients using Piezography inks ? Is there banding ? I mean, any banding at all ?

Better still, are there other ways to get a truly linearized profile ? I mean... linear.

Here's a 16-bit Grayscale TIFF file which reveals banding when present: http://www.kennethleegallery.com/images/forum/KenLeeBullsEye51.tif. It combines both a continuous gradient and 51-step wedge as a Bull's eye.

[Peter De Smidt]

You might try adding a bit of monochromatic Gaussian noise to the areas where banding occurs, just enough to eliminate the banding but not be visible otherwise.

[Ken Lee]

You might try adding a bit of monochromatic Gaussian noise to the areas where banding occurs, just enough to eliminate the banding but not be visible otherwise.

I want the print to look smooth, just as the image looks smooth on my calibrated monitor.

It's ironic that the QTR allows us to print in 16-bit mode, but (if I understand correctly) the actual profiles contain a lookup table of only 256 values (8-bit).

Meanwhile, if we measure only 21 steps, that's less than 5-bit precision. Am I missing something ? Is the rest handled with interpolation ? If so, is that really 16 bits ?

[Peter De Smidt]

With the right amount of noise, the image should still look smooth. I don't know the answers to the QTR questions. You might ask Roy.

[WalkerBlackwell]

There is no banding with Piezography inks. Partly its in the ink chemistry, partly in how we build the ink overlaps for the .quads that are built for Piezography, and partly its how we linearize.

All together, these three markers make perfectly smooth prints that exceed every other monochromatic system (besides maybe our most recent PiezoDN dig-neg system).

best,
Walker

[Ken Lee]

http://www.kenleegallery.com/images/forum/Noise1.jpg

Here's a photo made with a portrait lens. In the negative and scanned image viewed on the monitor, there is a smooth halo around each of the flowers. On my QTR print, the halos have 3 bands.

http://www.kenleegallery.com/images/forum/Noise2.png

Are you suggesting that I add Gaussian Noise to the whole photo, or just the halos ? Would I paint it in using a mask ? There are a lot of gradients to deal with.

[WalkerBlackwell]

Adding noise essentially negates the qualities of dedicated monochrome ink systems (my opinion only).

You might as well go with K3 at that point as this is essentially what K3 does (by spacing the dots of darker ink to make lighter tones) to mask defects in the linearization (or media-types).

When properly profiled, a qtr Piezography K6 or K7 print has significantly more resolution and fidelity than either K3 or a noised image . . . that is kinda the point.

best,
Walker

[Bruce Watson]

Prints of ordinary subjects look perfect. This problem only appears in images which contain substantial gradients. This isn't about print head alignment or cleaning cycles, this is about the quality of the profile.

Yes.


Somewhere I read that Piezography profiles are custom-linearized to exceed the imprecision inherent with ordinary QTR tools. This might explain why they are available for only a limited selection of papers.

Yes again.

I think you understand the problem and the solution.

That said, I've got a number of prints using the old 4 ink system that have smooth beautiful skies that are gradient city without a trace of banding. But as has been pointed out, this was done using some artful dithering. The six+ ink systems are able to resolve more detail in part because they can do away with this dithering, but at the cost of harder to make profiles (http://www.piezography.com/PiezoPress/tag/custom-profiles/).

[Peter De Smidt]

Possibly try the various dithering options in QTR. This is from a long time ago, but perhaps it is relevant: http://photo.net/digital-darkroom-forum/00BR5g

It does seem to be a profile issue, and coming up with a really good one is the most difficult thing about QTR.

For adding noise, I'd start by adding it to the whole print to see if the banding goes away while not adding any roughness. Assuming it works, I'd use a luminosity mask to pick the appropriate areas, using Quick Mask to paint out any problems.

If I were you, I'd send an email to Paul Roark asking for his suggestions.

[sanking]

Ken,

There should not be any banding in printing 16-bit files with either Piezography or with any other all gray ink set with a well-designed profile. Are you using the GRAY_OVERLAP command in your profile? If not, you might want to try it as I have found this feature very important for profiles where you have 6-8 gray shades. You might ask Richard Boutwell about this, as I recall he mentioned at some time in the past and had some suggestions for setting gama values in conjunction with the GRAY-OVERLAP command.

Sandy

[Ken Lee]

Hi Sandy - I have worked with Richard Boutwell quite a bit. He's a great guy. The best profile I have made, was done with his help.

In the end however, I was unable to entirely remove banding. Ultimately (I'm paraphrasing) he suggested that I was getting about as good a profile as can be expected with the tools at hand. I took this to mean that any further refinement would involve adjustments to the final .quad file - which would require a higher grade of measuring tools and mathematics than I possess.

The banding I'm describing is not blatant or apparent throughout the tonal scale: it appears in only a few places. Most of my prints look lovely and after working with Richard Boutwell I thought I'd give it a go with some of my most demanding images, like this one (http://www.kenleegallery.com/images/forum/Noise3.jpg): except for a few sharp edges, it's nothing but gradients.

On a related note, when I contacted the makers of my spectrophotometer in Switzerland, they told me that its inability to deliver a consistent read-out from the same sample was "normal". I could understand if we were talking about measurements down at the level of quantum uncertainty, but a simple LAB reading from a sheet of paper shouldn't be rocket science.

[Ken Lee]

I still wonder: if .QUAD files specify only 256 values for each channel (8 bits), how can we expect 16-bit fidelity ?

[Ken Lee]

Update: I made up a test image consisting of a the aforementioned 51-step bulls-eye gradient (http://www.kennethleegallery.com/images/forum/KenLeeBullsEye51.tif) and a few other challenging photos - then ordered a print from Cone Editions, using their pure Carbon inks and 100% rag paper.

The print is perfectly linear: no bumps, no banding, no blemishes of any kind. Perfect.

End of story for me.

[Peter De Smidt]

Ken, how's the print tone?

[Ken Lee]

I'm very happy with the print tone. It's warm, but I like it for most of my subjects.

They made the print on an Epson 3880 with K7 Carbon inks on Hahnemuhle Photo Rag 308gsm - a paper I have not previously used.

It's very similar to what I get on Premier Smooth Hot Press Fine Art Paper.

If perfect linearization weren't enough, the other factor to consider with InkjetMall and Cone Editions is their superb and prompt customer service: the difference is truly staggering.

[Peter De Smidt]

Ken, thanks for the feedback. I may make the plunge myself this fall.

[Ken Lee]

Additional update with further good news:

I ordered custom K7 Carbon profiles for my printer for 2 papers. They were made and emailed back to me in very quick time.

I just made test prints using these profiles and the results are just like the sample print they sent me: perfectly linear.

My old approach was to use a consumer-grade spectrophotometer, measure 21 steps and interpolate the points. Basing a profile on 21 measurements amounts to little more than 4-bit accuracy, so it's no wonder there was banding and irregularity which could not be adequately smoothed.

I found it's much more effective to let experts use professional tools to measure 256 steps and linearize.

Now I can finally make faithful prints of images like this one:


http://www.kennethleegallery.com/images/catalog/pottery/6.jpg

[sanking]

Ken,

Beautiful print. Congratulations on finding a solution that allows you to make such great prints from these delicate images.

Sandy

[williaty]

To clarify, you printed out a file Cone sent you, sent it back to them as a print, they read the print, and sent you profiles?

[Ken Lee]

Yes. You download a 256-step target, then print it using one of their existing profiles. You send them the print and the profile, they measure the print, adjust the profile to exactly suit your printer, and send you the profile.

Apparently they are in the process of developing a tool-set which will allow you to do this yourself, provided you use one of spectrophotometers they support.

[Christopher Barrett]

You can also make your own curves and profiles using QTR and Curve Calculator III (http://www.precisiondigitalnegatives.com/). Good densitometers aren't too expensive on eBay. I happen to have an i1 Pro 2, though and may use that.

[Ken Lee]

Is Curve Calcluator III designed strictly for making digital negatives for alt-process printing, or does it also work for direct printing ?

I notice that the author recommends the use of a 31-step target: "A 31-step provides increased accuracy in calibration over the 21-step tablet.". Is that the basis for calibration ? It's better than 21, but using 256 steps seems more promising to me on the face of it.

[WalkerBlackwell]

At IJM we calibrate at 256 steps but in 64bit computational space. It's complicated but the result is the result. Everything else out there is simply sub-par and noticeably so when you visually look at one print against another.

That said, we have recently optimized Roy's QTR-Linearize-Quad workflow to allow (DIY) iterative profiling that enables near perfect (smooth) linearization with an iterative method. These targets and optimization tools will be available soon. It will be available to Piezography customers.

Both PiezoDN digital negatives and print positives can be calibrated like this with Piezo inks.

regards,
Walker
R&D InkjetMall

[Richard Boutwell]

I know this is really late, but I wanted to chime in with a little more detail about why I thought Ken was getting the most he could with the tools at hand.

1. SpyderPrint is not accurate or consistent enough to do the kind of calculations needed to arrive at perfectly smooth k6-k7 profiles. The i1 Pro or one of the automated chart readers do better with this because they sample at 100-200 times a second depending on the model and software running it.

2. Many people follow the standard QTR workflow for setting the ink limits at a default limit for each ink. In my experience this does not give as even coverage as determining the density for each ink to evenly partition the gray scale. This is to prevent the inks from ramping up and down too quickly or using too many inks bunched up in the shadows and not having as much coverage throughout the highlights and midtones.

3. People are not accurately setting the cross over points—that is either because of guessing, following some bad advice, too aggressive rounding, or inaccuracies with the measurement device. Ken was doing the best he could here averaging several readings for interpolating the cross over points but I think there were still other problems that were out of his control. The increased gray overlap Sandy referenced can prevent or hide some of those errors but...

4. The biggest hurdle is not the tools or the calculations, but the way the QTR curve creation program changes the "shape" of the ink curve as new inks are introduced and the way they overlap. This is where the Piezography curves are really doing their magic. The bezier path that creates the long gentle slops and trailing edges in the Piezography curves hide any irregularities better than the QTR generated curves. I was experimenting with making bezier curves and then linearizing them with my own linearizer for Ken, but things got too busy and ended up going with the Cone inks and new carts (which are better anyway).

5. "everything else out there is sub par" for MOST people doing stuff with the standard QTR tools or hand mixed inks, yes.

[sanking]

I know this is really late, but I wanted to chime in with a little more detail about why I thought Ken was getting the most he could with the tools at hand.

1. SpyderPrint is not accurate or consistent enough to do the kind of calculations needed to arrive at perfectly smooth k6-k7 profiles. The i1 Pro or one of the automated chart readers do better with this because they sample at 100-200 times a second depending on the model and software running it.



I am curious why the sampling rate makes a difference in comparing result from a Spyder Print and iOne. Since the device is measuring a sample step one would expect that the "best" return would be an average of the entire step, or as large a circle of it as possible.

Recently I compared the results of linearizing a 51 step wedge with the iOne and a densitometer. When I compared the .quad file I noticed that the individual ink curves that resulted from linearizing the profile with the densitometer, which measures a circle of about 1/4" in diameter, were much smoother than the curves produced by scanning with the iOne.

In my case I assumed that the difference was due to small uneveneness in density within each step that result from hand coating and hand sensitizing with my carbon transfer process. But that in turn got me to thinking about how the various devices actually return readings for a single step from the step wedge. I know that the return from the densitometer is an average of the 1/4" aperture, but am not sure how this is computed by the iOne or Spyder Print.

Sandy

[Richard Boutwell]

I am curious why the sampling rate makes a difference in comparing result from a Spyder Print and iOne. Since the device is measuring a sample step one would expect that the "best" return would be an average of the entire step, or as large a circle of it as possible.

I can not say for certain, but I think the small variations in measurement angle from pressing the button on the spyderprint and single patch measurements is what is causing the dramatic sample to sample differences. I don't have much experience with different densitometers, but the x-rite i have is super solid and the measurement angle does not change at all.




Recently I compared the results of linearizing a 51 step wedge with the iOne and a densitometer. When I compared the .quad file I noticed that the individual ink curves that resulted from linearizing the profile with the densitometer, which measures a circle of about 1/4" in diameter, were much smoother than the curves produced by scanning with the iOne.

There are a few questions I have been chewing on for a while. Most Density measurements are not as precise as L* measurements but L* might show more errors. So, you would think there would be a benefit of measuring in Density because the errors basically get rounded off. The problem is that any precision gets rounded off too... So is it better to average off the errors in the L* measurements by simply routing off, or averaging several passes of from an i1, or the simple moving mean error correction the piezoDN/peizoPro system uses?

Then there is also the issue of the quality of the measurement device and how it records and transmits the signal back to the software. Not being an electrical engineer I can really only make assumptions, but higher quality devices might have better voltage stabilization in the unit to provide more consistent illumination and recording of each sample (and that is where more samples per patch in scan mode would conceivably be better than the single spyderprint measurements). Unless you are using a voltage stabilizer and an automated chart reader like the dtp-41 or 70 most handheld devices get power over USB, and there might be slight fluctuations that cause the errors from one patch to the next.

The other main issue with the resulting quad curves is that when you are linearizing with the QTR tools the Density measurements are translated to xyz_y then to Lab_L*. You might see more bumps in the linearizing from the original L* values but not in the quad curves from original Density measurements because the errors are rounded off to 2 decimal places before being converted to xyz_y to L*. Basically the number of decimal places in the Density measurements might not show the errors in the precision of the L* measurements. The issue is balancing precision with the degree of error and deciding if/how to smooth the errors...



In my case I assumed that the difference was due to small uneveneness in density within each step that result from hand coating and hand sensitizing with my carbon transfer process. But that in turn got me to thinking about how the various devices actually return readings for a single step from the step wedge. I know that the return from the densitometer is an average of the 1/4" aperture, but am not sure how this is computed by the iOne or Spyder Print.

Sandy


I had the same issue when I was building my pre-piezoDN linearizer earlier in the year. At first I thought my correction curve math was bad because I had regular bumps in the resulting measurement file (but not in the gradient) but It turned out to be inconsistent coating between the top of the chart and the bottom (The simple moving average was enough to solve that, and it was interesting to see that is what Walker had done with the CGATS smoother tool also). This data smoothing thing wasn't a problem with my inkjet print linearizer, where I was just using averages from 3 51-step targets from the i1. However, when linearizing from a larger patch chart, like those with 129-256 patches, it does require some data smoothing no matter how good the measurement device or the number of samples per patch. That might be caused by variations in the paper surface, slight changes in the measurement angle, drying time, temperature, humidity, tidal forces...

[Ken Lee]

Measuring a sample with my Spyder spectro - without moving the sensor at all - I get the same Density reading every time. Switching to Lab, I get a different reading every time. I complained to the manufacturer about it and they told me it was normal. Whether its voltage instability or some other factor, I figured it's an inferior device.

Whatever the instrument, I realized that sampling only 21 steps is inadequate for fine art printing, being just over 4-bit precision. Below is a section from the piezography target "proofofpiezography.tif". It contains sections from the low and high end of the tonal scale in 1% increments. My calibrated iMac can barely separate the high values, and completely blocks the low value. However, my piezography prints clearly distinguish each step. 100 steps is less than 7-bit precision, still shy of 8-bit and far below 16-bit precision. My piezography prints distinguish all 255 steps, giving real 8-bit precision.

For me, now the problem is that the Piezography system has raised the bar of tonal fidelity to the point where I need to consider getting a real monitor like an NEC Spectraview or Eizo which can see the entire tonal scale. Simple monitor profiling software doesn't get us there, particularly at the low end of the tonal scale. See The K7 standard and monitor display systems (http://piezography.com/2008/12/27/the-k7-standard-and-monitor-display-systems/), written all the way back in 2008. Perhaps if we consider the paper wasted in making too many proofs, such a monitor will pay for itself.

http://www.kennethleegallery.com/images/forum/piezotarget.png

[WalkerBlackwell]

At Cone Editions and InkjetMall we have a unique algorithm for fixing inherent measurement errors in spectrophotometers that is not an average of multiple readings or a line-average of an array of different readings on a scale. We have updated this algorithm and it will be available for single-seat use within a tool called the "Piezography Error Corrector" as part of the Piezography Professional Edition toolset for 51 and 129 step targets. We'll be launching in 45 days or so. info at our newly re-vamped website http://piezography.com

best,
Walker