A new Printer/Media Resolution test target

[Bart_van_der_Wolf]

Hi Bart, thank you for the targets. These are brilliant. Previously I used your camera resolution target to check alignment and positioning accuracy for my printer.

Hi Samuel, you're welcome. Yes, the camera resolution target is also somewhat usable for testing the printer quality, because it has some checks built in to test for good output quality before shooting it with a camera. It wouldn't make sense to test a camera if the test target is no good.

The single pixel lines are super taxing on the printer. My Canon iPF8400 can resolve both the horizontal and vertical lines perfectly when it first arrived, but it underwent a carriage assembly replacement and now the vertical lines (perpendicular to the direction of head travel) are not as crisp or well defined as the horizontal lines. The patch looks a little more dense, if memory serves.

The print has a known (presumably non-resampled) output size, unlike shooting a target with a camera that will vary the on sensor image magnification with focal length and shooting distance, and add blur due to defocus and diffraction. That's why I could up the ante for the print resolution target detail, and add fixed/discrete resolutions with known physical characteristics. That also made it much easier to visually compare how much ink diffusion blurs detail and adds density due to dot gain. The influence of ink/media combinations on resolution becomes better visible, and even quantifiable.

Another use for the target is then to actually demonstrate how viewing distance will affect the resolution requirements. With a good print of the target (even at a magnified output size), one can view it at different distances, and conclude that at a larger distance one can print larger without visual penalty (assuming proper resampling and output sharpening). Being able to visually resolve the 8 cycles/mm patch detail, indicates excellent output resolution, but much harder to even see than the already good 5 cycles/mm detail.

I am working on other things at the moment but I would love to try out the new target soon. Getting the printheads to be perfectly aligned is surprisingly difficult. I have tried to run several manual and auto (including the intitial setup adjustment, which is the most thorough) and realised that the initial head adjustment is the best and most likely to be successful for my iPF8400. The manual adjustment was too unpredictable and dodgy, and does not print adjustment patterns for yellow and one other color I think, because I suspect it would be too faint to see. There usually still is some minor positioning errors over the page, especially in the initial few inches of the paper because of the curl from the roll.

Indeed, perfect alignment is a tough goal to achieve. This target is very sensitive, so I wouldn't be surprised if some minor issues remain detectable. But I prefer to aim for a very high goal and almost reach it, than to meet lower standards with relative ease. As long as perfection doesn't become an obsession, it's worthwhile to keep aiming for it.

I have seen a definite resolution difference between 300ppi and 600ppi data sent to this printer. Ming Thein has also shown a big improvement going from 360 ppi to 720 ppi on the Epson 9900 on his blog.

This is great, can't wait to try it! Thank you.  

Yes, hopefully by now, most people are convinced that there is a real benefit beyond 300 or 360 PPI, although it is not equally beneficial for all image content. Also, output sharpening remains an often undervalued asset, and having more pixels allows to do more. There might also be some wide format output limitations as to how many pixels per paper width can be fed to some printer drivers.

Cheers,
Bart

[azmike]

Thank you Bart for developing and sharing the targets.   I think many of us have long wondered "How good is my print (compared to how good it could be)?" Your target is a very useful reference.

When I first printed the 600 ppi target on my HP Z3200 I was more than disappointed....lots of aliasing and banding (that really prevented any reasonable interpretation of printer resolution performance).  After two head alignments and paper advance calibrations, and tinkering with the "best/max detail/more passes" settings the printed target hadn't improved.

And then I went back to the target image size as reported by CS6 that was "5.118 inches" at "599.999 pixels/inch" and reset the resolution to "600.000 pixels/inch" (when I'm using a "reference" the first thing I try not to do is to start adjusting the reference).  Now the target printed very nicely (without the strange aliasing and banding).  The 7 cycle/mm patch prints nearly perfectly.

Thanks again for your longstanding and many contributions to LULA.

Mike Coffey
Prescott, AZ

[Bart_van_der_Wolf]

Thank you Bart for developing and sharing the targets.   I think many of us have long wondered "How good is my print (compared to how good it could be)?" Your target is a very useful reference.

Hi Mike,

You're right, it's perhaps more useful to get a confirmation of what the situation actually is, instead of how it might be in theory.

When I first printed the 600 ppi target on my HP Z3200 I was more than disappointed....lots of aliasing and banding (that really prevented any reasonable interpretation of printer resolution performance).  After two head alignments and paper advance calibrations, and tinkering with the "best/max detail/more passes" settings the printed target hadn't improved.

You had me also worried as I was reading this up to here ...

And then I went back to the target image size as reported by CS6 that was "5.118 inches" at "599.999 pixels/inch" and reset the resolution to "600.000 pixels/inch" (when I'm using a "reference" the first thing I try not to do is to start adjusting the reference).  Now the target printed very nicely (without the strange aliasing and banding).  The 7 cycle/mm patch prints nearly perfectly.

Phew, I'm relieved. I told it was sensitive to small alignment errors, but this is a good warning to double-check the print settings before questioning other things in the setup. And let's face it, 599.999 PPI is not the same as 600 PPI.

Anything better than 5 cycles/mm is exceeding 254 PPI and is considered good quality, so 7 cycles/mm which equals 355.6 PPI,  exceeds the 300 PPI limit which proves that more resolution is to be had with the right driver settings. Even better quality, although sometimes very hard to see, is 8 cycles/mm or 406.4 PPI. It may not be achievable with all papers, and maybe not on all printers. It would take very close scrutiny anyway to see it.

Thanks again for your longstanding and many contributions to LULA.

It's my pleasure. I also pick up some useful info here and there, so we all get to benefit by sharing. I wouldn't have known there was a use for such a tool if it weren't for the discussions about print resolution. It wasn't until I came across some print samples that exhibited significant ink diffusion that I wondered what the actual resolution limit was for such materials, and if 600 or 720 PPi was even needed for those. Now we can tell, and also if head alignment or paper travel is holding us back, in which case perhaps some additional cleaning of the transport path or mechanism might help.

Cheers,
Bart

[samueljohnchia]

Bart, I spent this afternoon using your target to evaluation the alignment accuracy of my Canon iPF8400 printer, after a recent 'operation' to a mechanical component. I noticed something immediately - aliasing artifacts in the print. Like Mike, I discovered that the 600ppi target was set to 599.999ppi instead of 600ppi. I resized without resampling to 600ppi and printed again, this time the 'aliasing artifacts' were still present but looked different. Upon closer inspection of the target (digital file) itself on my monitor, I noticed that the cycles/mm lines are not all of equal contrast gradients - the bands I see in the prints correspond to the bands in the target. So my printer was accurately resolving the artifacts in the target afterall. Could you confirm if my observations on this is correct? In your OP you said the target should be virtually free of defects.

I used the Canon 16 bit PS Plug-in to make my prints, on Harman by Hahnemuhle Gloss Baryta. Depending on the media setting used, the Plug-in does not allow for 300ppi input resolution, only 600ppi resolution. Most of the media settings intended for printing on high quality papers have the 300ppi option grayed out. I decided to make two reference prints, one for the 'Highest (max no of passes)' setting and one for 'Highest". The former used to be 32 passes on the X1XX printers, and now is 16 passes on the X3XX and X4XX printers. The latter I believe is 7 passes, used to be 8. I have done this experiment several times before, and have long known about the much smoother dither and more uniform coverage achieved using the 'Highest (max no of passes)' setting. There is a small but slight resolution increase as well. The vertical resolution is somewhat lower, probably to the limitations of the printer, unable to as accurately fire a vertical column of dots as the paper advances over more passes. Horizontal lines are cleaner however, with less overspray effect.

Here is a link to a zip folder containing 4 jpegs of actual size macro photographs of the prints themselves. The magnification is close to 1:1, taken using a Canon 100mm macro lens on a 5D II.

https://www.dropbox.com/s/i0iqzx1qaqj8bdj/Printer%20Resolution%20Target%20comparison.zip?dl=0

Another use for the target is then to actually demonstrate how viewing distance will affect the resolution requirements. With a good print of the target (even at a magnified output size), one can view it at different distances, and conclude that at a larger distance one can print larger without visual penalty (assuming proper resampling and output sharpening). Being able to visually resolve the 8 cycles/mm patch detail, indicates excellent output resolution, but much harder to even see than the already good 5 cycles/mm detail.

Yes, it is very useful for that! What a brilliant way to quantify resolution at viewing distances. To my eyes, the smoother dither of the 'Highest (max no of passes)' setting is visible to about 15 inches away. The higher resolution (cleaner, crisper lines) in the 4 cycles/mm lines are visible to about 3 feet away.

There is a huge speed penalty however, prints take roughly twice as long to print using this setting. For example, a 24 x 36 inch print takes about 30 mins to complete, a long time but I dearly love the quality improvement, which is visible even on textured matte fine art papers.

[Bart_van_der_Wolf]

Bart, I spent this afternoon using your target to evaluation the alignment accuracy of my Canon iPF8400 printer, after a recent 'operation' to a mechanical component. I noticed something immediately - aliasing artifacts in the print. Like Mike, I discovered that the 600ppi target was set to 599.999ppi instead of 600ppi. I resized without resampling to 600ppi and printed again, this time the 'aliasing artifacts' were still present but looked different. Upon closer inspection of the target (digital file) itself on my monitor, I noticed that the cycles/mm lines are not all of equal contrast gradients - the bands I see in the prints correspond to the bands in the target. So my printer was accurately resolving the artifacts in the target afterall. Could you confirm if my observations on this is correct? In your OP you said the target should be virtually free of defects.

Hi Samuel, great question, which may have been puzzling others as well, but they may have been afraid to ask... Glad you did.

One part of the difficulty is related to the high level of detail that approaches the Nyquist frequency of the printer. To reproduce a high spatial frequency unambiguously, we need more than 2 pixels per cycle. The 2 px/cy is reached by some of the patterns (top left, and near the green circle) but it is still very sensitive to small irregularities.

The difficulty of totally avoiding aliasing most likely has to do with gamma. When you look at the target on your (LCD, because CRT will create other issues) display at 100% zoom, AND that display is perfectly calibrated for Gamma 2.2, it will be possible to see aliasing IF one moves the observing position a bit up or down, or when viewing the display at a non-perpendicular angle. The patterns are calculated to be as perfect as possible in discrete pixels, but also needs a bit of averaging of intermediate tones that actually have multiple levels of brightness across a single discrete pixel. That averaging is calculated to be most neutral at a display Gamma of 2.2.

But even at that level of physical constraint, even the viewing position/angle apparently is able to introduce aliasing. That's how sensitive the patterns are. So it is not surprising that printing, with additional profile conversions and ink diffusion and media surface structure and dithering, will also be hard/impossible to reproduce flawlessly. The goal should therefore be to minimize the flaws, like density difference between horizontal and vertical running cycles. Luckily for us, normal images do not pose such challenges, or at least we cannot see them as clearly (the issues are still there, hidden by variable image detail).

Maybe, the best possible angle of attack would be to assign the output profile to the target and then print it with that same profile (if the OS doesn't interfere), to reduce the risk of gamma conversions between profiles / colorspaces. But I would still expect a certain level of aliasing to occur. Do also note that e.g. Adobe's ACM adds a linear slope to an AdobeRGB tagged colorspace in images in the lowest levels, so even that is not a uniform gamma space under that regime.

You can also, by simulation, learn to detect the difference between gamma induced aliasing, and that caused by dimensional influences (although the dithering patterns may also have an influence). When you (e.g. in Photoshop) temporarily change the overall gamma (in Levels, or with a curves adjustment) of the target (don't save!), you can see that mostly the horizontal/vertical running cycles will show aliasing. The diagonal running cycles can use a 41% higher accuracy of detail placement and are less affected due to being already somewhat gamma averaged or contrast reduced. However, when changing the image size a bit by resampling, the diagonal running cycles will attract a square aliasing pattern, consistent with a higher diagonal resolution capability. In a dithered print these effects may look a bit different, but it's possible that similar effects can be detected.

I used the Canon 16 bit PS Plug-in to make my prints, on Harman by Hahnemuhle Gloss Baryta. Depending on the media setting used, the Plug-in does not allow for 300ppi input resolution, only 600ppi resolution. Most of the media settings intended for printing on high quality papers have the 300ppi option grayed out. I decided to make two reference prints, one for the 'Highest (max no of passes)' setting and one for 'Highest". The former used to be 32 passes on the X1XX printers, and now is 16 passes on the X3XX and X4XX printers. The latter I believe is 7 passes, used to be 8. I have done this experiment several times before, and have long known about the much smoother dither and more uniform coverage achieved using the 'Highest (max no of passes)' setting. There is a small but slight resolution increase as well. The vertical resolution is somewhat lower, probably to the limitations of the printer, unable to as accurately fire a vertical column of dots as the paper advances over more passes. Horizontal lines are cleaner however, with less overspray effect.

Yes, there is also a potential effect due to the droplets being fired at the medium while having a horizontal speed, thus creating a bit of a splatter pattern instead of a purely round drop, depending on the efficiency of the absorption of the top coated layer. So a slower travel speed may give slightly rounder droplet patterns, and unidirectional can look a bit different from bi-directional (in addition to hysteresis in positioning accuracy).

Here is a link to a zip folder containing 4 jpegs of actual size macro photographs of the prints themselves. The magnification is close to 1:1, taken using a Canon 100mm macro lens on a 5D II.

https://www.dropbox.com/s/i0iqzx1qaqj8bdj/Printer%20Resolution%20Target%20comparison.zip?dl=0

Thanks for those. I was planning to post some results, but I'm waiting for the profiler patterns to dry before creating another profile for a printer I'm testing. So you beat me to it ...

The 'Highest (max no of passes)' does show a subtle (but clearly visible) improvement on the limiting resolution, and a much smoother rendering of the star region. Of course we're pixel peeping, and comparing side-by-side, but there is an obvious benefit for improved quality.
The 8 cycles/mm patch looks pretty good, with almost the same density for hor/ver running cycles. The aliasing pattern, suggests a bit more gamma influence than dimensional cause to the aliasing pattern. That seems to be relatively common.
The smoothness is markedly improved, text is cleaner, and the gradient looks better (very well profiled, all the way to black).
The top left patterns demonstrate very good uniform alignment of both print-head as wel as paper travel direction. Again the 'Highest (max no of passes)' setting proves to be superior in resolving fine detail with better smoothness.

Yes, it is very useful for that! What a brilliant way to quantify resolution at viewing distances. To my eyes, the smoother dither of the 'Highest (max no of passes)' setting is visible to about 15 inches away. The higher resolution (cleaner, crisper lines) in the 4 cycles/mm lines are visible to about 3 feet away.

The resolution requirements change linearly with distance (assuming well functioning eyes/correction). So it will be simple to test for different distances. Of course some (often younger) people have better visual acuity than others.

There is a huge speed penalty however, prints take roughly twice as long to print using this setting. For example, a 24 x 36 inch print takes about 30 mins to complete, a long time but I dearly love the quality improvement, which is visible even on textured matte fine art papers.

Yes, speed takes a hit, but another of those urban myths that needs to be dispelled, is that all matte media or even canvas surfaces are blurry. There are huge differences ...

Cheers,
Bart

[Bart_van_der_Wolf]

One more thing to clarify.

Upon closer inspection of the target (digital file) itself on my monitor, I noticed that the cycles/mm lines are not all of equal contrast gradients - the bands I see in the prints correspond to the bands in the target. So my printer was accurately resolving the artifacts in the target afterall.

I've designed the target as a relatively 'cruel but honest' one, but it is possible to design a 'gentler for softies' version.

To illustrate, I've attached 3 versions of the 8 cycle/mm (horizontal cycles patch area) for 600 PPI printers (next post will have the three for 720 PPI). They have exactly the same spatial frequency, but they differ in the amplitude due to added anti-aliasing. The anti-aliasing may make the target patch look more benign, but it also makes it harder to detect the higher resolution patches resolutions. That will also affect the ability to test potential future print technologies with higher resolutions, and be less sensitive to show resampling artifacts.  So the question becomes, which is more useful, gentle or strong (I can create any intermediate level)?

My current take is that, as long as we know what can happen with regard to aliasing, it is beneficial for the test to be relatively sensitive in order to more clearly show small improvements/degradations.

Cheers,
Bart

[Bart_van_der_Wolf]

Here I've attached 3 versions of the 8 cycle/mm (horizontal cycles patch area) for 720 PPI printers.

Cheers,
Bart

[samueljohnchia]

One part of the difficulty is related to the high level of detail that approaches the Nyquist frequency of the printer. To reproduce a high spatial frequency unambiguously, we need more than 2 pixels per cycle. The 2 px/cy is reached by some of the patterns (top left, and near the green circle) but it is still very sensitive to small irregularities.

The difficulty of totally avoiding aliasing most likely has to do with gamma. When you look at the target on your (LCD, because CRT will create other issues) display at 100% zoom, AND that display is perfectly calibrated for Gamma 2.2, it will be possible to see aliasing IF one moves the observing position a bit up or down, or when viewing the display at a non-perpendicular angle. The patterns are calculated to be as perfect as possible in discrete pixels, but also needs a bit of averaging of intermediate tones that actually have multiple levels of brightness across a single discrete pixel. That averaging is calculated to be most neutral at a display Gamma of 2.2.

But even at that level of physical constraint, even the viewing position/angle apparently is able to introduce aliasing. That's how sensitive the patterns are. So it is not surprising that printing, with additional profile conversions and ink diffusion and media surface structure and dithering, will also be hard/impossible to reproduce flawlessly. The goal should therefore be to minimize the flaws, like density difference between horizontal and vertical running cycles. Luckily for us, normal images do not pose such challenges, or at least we cannot see them as clearly (the issues are still there, hidden by variable image detail).

Hi Bart, thank you for your detailed reply. I think I know what you mean. Here are two crops with an overlay of little red arrows pointing to some of the effects that I saw in the target which appeared like aliasing, which as you describe, is due to the inherent difficulty of using very few discrete pixels to contain a sinusodial pattern.:





While I agree that the gamma is an issue to beware of, zooming to 400% or more and looking at the pixels themselves show the diagonal lines in the 8 cycles/mm box to be darker on average than the horizontal or vertical pixels. When I first saw the print, I mistakenly thought the 4 quadrants were supposed to print with the same average brightness, until I studied the target file.

Regarding the design of the target, I realise that I am really only looking closely at just two parts of the target when using it to evaluate the positioning accuracy of my printer - the single pixel lines in the top left, and the smooth gray background to check the smoothness/evenness of the dither. I'm now wondering how useful the cycles/mm strips are since they look so funny with the aliasing artifacts. What if the patterns are composed of uniform contrast lines instead of sinusoidal lines? Of course they would not be in the same cycles/mm as the current target.

Maybe, the best possible angle of attack would be to assign the output profile to the target and then print it with that same profile (if the OS doesn't interfere), to reduce the risk of gamma conversions between profiles / colorspaces. But I would still expect a certain level of aliasing to occur. Do also note that e.g. Adobe's ACM adds a linear slope to an AdobeRGB tagged colorspace in images in the lowest levels, so even that is not a uniform gamma space under that regime.

I did assign the output profile instead of converting to it, but also tried conversion as well, and could see no real significant difference in the aliasing pattern, except that the target looked much more neutral when converting to printer profile.

Yes, I am aware that Adobe adds a linear slope to all gamma encoded working spaces, not just Adobe RGB. It helps against numbers getting too large as the slope of the curve approaches vertical.

You can also, by simulation, learn to detect the difference between gamma induced aliasing, and that caused by dimensional influences (although the dithering patterns may also have an influence). When you (e.g. in Photoshop) temporarily change the overall gamma (in Levels, or with a curves adjustment) of the target (don't save!), you can see that mostly the horizontal/vertical running cycles will show aliasing. The diagonal running cycles can use a 41% higher accuracy of detail placement and are less affected due to being already somewhat gamma averaged or contrast reduced. However, when changing the image size a bit by resampling, the diagonal running cycles will attract a square aliasing pattern, consistent with a higher diagonal resolution capability. In a dithered print these effects may look a bit different, but it's possible that similar effects can be detected.

I just realised that ! cannot resize you 600ppi target to 599.999ppi in Photoshop! There is no change to the pixel resolution at all. So I have no idea what Mike was seeing. The aliasing pattern in my first bad print was my printer being totally out of alignment initially - there was a mechanical adjustment to the paper feed system and I had not run an alignment yet.

Cool tip about playing with the gamma slider using a levels adjustment! I like that, seeing the aliasing patterns get more or less visible is fun. My display gamma is definitely not a perfect 2.2 (whose is?) and there are slight gamma induced (or tone curve induced) changes, making the aliasing a little more visible than on screen.

Thanks for those. I was planning to post some results, but I'm waiting for the profiler patterns to dry before creating another profile for a printer I'm testing. So you beat me to it ...

Sorry to have stolen your thunder! Please post your samples too, I would like to know what you managed to achieve on your end.

And thanks for your observations! I now think the printer is capable of even better performance, after looking at some of my older prints. The new Canon iPFs have excellent built-in linearisation adjustments when performing the paper calibration, which can be paper specific if you have special custom media setting files (built using the iPF6450 with spectrophotometer). The LUTs for the X4XX printers according to Scott Martin have been updated to improve on linearity and smoothness slightly. The printer profile can only do so much if the printer could not separate well near black. In general I am quite happy with its performance, but like everything else, there is room for improvement. Btw I find the smoother vs rougher dither for the two highest quality modes to be more easily seen with the naked eye than in my macro photographs.

Yes, speed takes a hit, but another of those urban myths that needs to be dispelled, is that all matte media or even canvas surfaces are blurry. There are huge differences ...

That was exactly what's on my mind. I can see the smoothness advantage even on canvas, even though the resolution the media is capable of is slightly lower. I print all of my clients work at the highest quality setting, regardless of size, despite the huge efficiency penalty, because it just looks much better up close.

Yes, there is also a potential effect due to the droplets being fired at the medium while having a horizontal speed, thus creating a bit of a splatter pattern instead of a purely round drop, depending on the efficiency of the absorption of the top coated layer. So a slower travel speed may give slightly rounder droplet patterns, and unidirectional can look a bit different from bi-directional (in addition to hysteresis in positioning accuracy).

I would agree with that line of reasoning, but looking closely at the dots under 40x magnification, there is no splatter pattern around the dot? The dots are just not being fired as precisely as they should be, creating a blurry fuzz around what should be a clean edge. I have never been able to see the splatter from my printer's dots! Much too small. In my old tests I clearly observed issues with bidirectional printing affecting dot placement precision, affecting vertical line resolution and hurting the overall smoothness of the dither pattern.

Here is something curious. This is an image of the heading text from a banding adjustment pattern, which seems to be a 4 pass print mode or something like that, certainly not as many passes as Highest (max no. of passes). It may have been 7 passes. The text is much cleaner looking at the edges, proof that my iPF8400 is capable of very clean looking lines. For some reason when there is a lot to do in a pass, printing highly varying detail, over an increased number of passes using the highest quality mode, the dot precision gets all messed up when it comes to clean edges like text.

I am looking again at some of my older prints of your camera resolution chart in the single/double pixel vertical and horizontal lines and noticing cleaner lines, but less accurate dot placement. Like I said in an earlier post, printhead and paper feed alignment is hard to get perfect! I have prints from an Epson 7900 that shows exceedingly smooth dither, but is unable to resolve single pixel lines (@720ppi) as well as the Canons can resolve them (@600ppi). Looks like I need to try aligning the heads yet again...



(right click and open in new tab to view at full size)

[samueljohnchia]

One more thing to clarify.

I've designed the target as a relatively 'cruel but honest' one, but it is possible to design a 'gentler for softies' version.

To illustrate, I've attached 3 versions of the 8 cycle/mm (horizontal cycles patch area) for 600 PPI printers (next post will have the three for 720 PPI). They have exactly the same spatial frequency, but they differ in the amplitude due to added anti-aliasing. The anti-aliasing may make the target patch look more benign, but it also makes it harder to detect the higher resolution patches resolutions. That will also affect the ability to test potential future print technologies with higher resolutions, and be less sensitive to show resampling artifacts.  So the question becomes, which is more useful, gentle or strong (I can create any intermediate level)?

My current take is that, as long as we know what can happen with regard to aliasing, it is beneficial for the test to be relatively sensitive in order to more clearly show small improvements/degradations.

Cheers,
Bart

Thanks for these additional patches. Yeah, the lower contrast bottom most patch is probably not as useful, and the artifacts are still quite visible and an educated guess is it will still show in the print. I guess I need to re-think how I use this target for evaluating the resolution of printer+media.

[Some Guy]

Anyone try this using Qimage Ultimate and Photoshop CS6?

I got "wildly" different results even though the images both measure out to 5.118" sq. and 720ppi.

Here is a scan of them both, but they are strange to say the least!  Same 3880 printer used.  Head alignment done yesterday too.

Don't know why PS see the upper right square 14.2 as a B&W only.  QI does better there.

I also did one on a 3880 with K7 Piezo inks and QTR.  Had to convert it to B&W Gamma 2.2 TIFF and it was better, although some shadows in the vertical line areas of the outside blocks.

Well crap!  Even Lightroom 5.7.1 prints differently than PS CS6 too.  No B&W square like the image shown below at 14.2 in PS CS6 (Which is the only one of 4 programs to do that!).

This is a cruel joke!  Even Corel PaintShop Pro X7 behaves differently, closer to Photoshop CS6 and with a different B&W square in the upper left near 14.2.

Added in Zoner Photo Studio 16 Pro x64 and it too is different in that the B&W squares are now all black! 

Pick your software.  They are all different.  Sort of like no two raw converters are the same.

SG

[Czornyj]

Like I said in an earlier post, printhead and paper feed alignment is hard to get perfect! I have prints from an Epson 7900 that shows exceedingly smooth dither, but is unable to resolve single pixel lines (@720ppi) as well as the Canons can resolve them (@600ppi). Looks like I need to try aligning the heads yet again...

Canon iPF is better at resolving fine details, so I'm afraid aligning won't help in this regard.

[Bart_van_der_Wolf]

Hi Bart, thank you for your detailed reply. I think I know what you mean. Here are two crops with an overlay of little red arrows pointing to some of the effects that I saw in the target which appeared like aliasing, which as you describe, is due to the inherent difficulty of using very few discrete pixels to contain a sinusodial pattern.

Correct, it only appears like that because we humans see patterns (like faces in moon surface rocks) even when they are not there. The patterns you correctly point at are merely limitations of pixel size, which aren't small enough to depict gradients accurately. Having said that, I am working on generating even higher quality patterns, but they take a lot longer to calculate (processing time x 100) and then need to be composited into the various composite patterns. May take a while to complete.

While I agree that the gamma is an issue to beware of, zooming to 400% or more and looking at the pixels themselves show the diagonal lines in the 8 cycles/mm box to be darker on average than the horizontal or vertical pixels. When I first saw the print, I mistakenly thought the 4 quadrants were supposed to print with the same average brightness, until I studied the target file.

Not mistakenly, they are supposed to be of equal density (especially the bottom two that run hor/ver), but gamma and profiling may play a trick on us, as well as the difficulty to exactly get intermediate tones from a limited number of inks through dithering. I do see the issue with aliasing patterns, the low frequency waves, and am looking for a way to solve that without compromising the detection sensitivity of errors. Maybe I need to change the overall gamma of the target, or use a different pattern than pure sinusoids (but that would probably need different settings for each dithering method). At the moment I'm calculating the patterns with much higher accuracy to see if that makes a difference. If not, more drastic changes may be needed.

Regarding the design of the target, I realise that I am really only looking closely at just two parts of the target when using it to evaluate the positioning accuracy of my printer - the single pixel lines in the top left, and the smooth gray background to check the smoothness/evenness of the dither. I'm now wondering how useful the cycles/mm strips are since they look so funny with the aliasing artifacts. What if the patterns are composed of uniform contrast lines instead of sinusoidal lines? Of course they would not be in the same cycles/mm as the current target.

Unfortunately, uniform contrast lines only allow fixed pixel resolution measurements, like 600 PPI = 11.81 cy/mm, 300 PPI = 5.91 cy/mm, 200 PPI = 3.94 cy/mm, 150 PPI = 2.95 cy/mm , 120 PPI = 2.36 cy/mm, 100 PPI = 1.97 cy/mm, and only in horizontal/vertical orientation. But those resolutions should be easy for any printer. It's mainly in the region between 300 or 360 PPi and 600 or 720 PPI that we will see the differences in e.g. media/ink diffusion, and there would be no info there without sinusoidal or other intermediate tones.

I did assign the output profile instead of converting to it, but also tried conversion as well, and could see no real significant difference in the aliasing pattern, except that the target looked much more neutral when converting to printer profile.

Good. Frankly, your profiling seems to be quite good.

I just realised that ! cannot resize you 600ppi target to 599.999ppi in Photoshop! There is no change to the pixel resolution at all. So I have no idea what Mike was seeing. The aliasing pattern in my first bad print was my printer being totally out of alignment initially - there was a mechanical adjustment to the paper feed system and I had not run an alignment yet.

One can change the pixel size of the target, e.g. by 1 pixel (or more) to reduce the PPI by lowering the  pixel count, yet print at the same output size. But that will activate the printer driver's resampling, which is of much lower quality than we can do. By only reducing the pixel count, and printing a proportional fraction smaller (keeping a constant 600 PPI), we will see the aliasing caused by Photoshop's resampling.

I would agree with that line of reasoning, but looking closely at the dots under 40x magnification, there is no splatter pattern around the dot? The dots are just not being fired as precisely as they should be, creating a blurry fuzz around what should be a clean edge. I have never been able to see the splatter from my printer's dots! Much too small. In my old tests I clearly observed issues with bidirectional printing affecting dot placement precision, affecting vertical line resolution and hurting the overall smoothness of the dither pattern.

I assume that the head travel speed is chosen to minimize any speed splatter on common media. It also ties into the dying/blending characteristics for multiple passes, so timing is probably carefully selected.

]Here is something curious. This is an image of the heading text from a banding adjustment pattern, which seems to be a 4 pass print mode or something like that, certainly not as many passes as Highest (max no. of passes). It may have been 7 passes. The text is much cleaner looking at the edges, proof that my iPF8400 is capable of very clean looking lines. For some reason when there is a lot to do in a pass, printing highly varying detail, over an increased number of passes using the highest quality mode, the dot precision gets all messed up when it comes to clean edges like text.

It's odd indeed. A long shot, but maybe a print from another application can make a difference? Could there be output dithering taking place by the Color Management system?

Cheers,
Bart

[Bart_van_der_Wolf]

Anyone try this using Qimage Ultimate and Photoshop CS6?

I got "wildly" different results even though the images both measure out to 5.118" sq. and 720ppi.

Here is a scan of them both, but they are strange to say the least!  Same 3880 printer used.  Head alignment done yesterday too.

Don't know why PS see the upper right square 14.2 as a B&W only.  QI does better there.

I also did one on a 3880 with K7 Piezo inks and QTR.  Had to convert it to B&W Gamma 2.2 TIFF and it was better, although some shadows in the vertical line areas of the outside blocks.

Well crap!  Even Lightroom 5.7.1 prints differently than PS CS6 too.  No B&W square like the image shown below at 14.2 in PS CS6 (Which is the only one of 4 programs to do that!).

Hi SG,

Thanks for that feedback. Maybe you're on to something, if Samuel is printing from PS (although at 600 PPI, so the frequency effects will be different).

Cheers,
Bart

[samueljohnchia]

Anyone try this using Qimage Ultimate and Photoshop CS6?

I got "wildly" different results even though the images both measure out to 5.118" sq. and 720ppi.

Here is a scan of them both, but they are strange to say the least!  Same 3880 printer used.  Head alignment done yesterday too.

I just downloaded your scan jpegs, they are quite small but still very revealing. I do not think it is a head alignment issue at all.

First up Bart's target for Epson (360/720ppi) printers is not 720ppi - it is 719.988ppi. You need to resize without resampling to 720ppi for one.

If you failed to catch that, I suspect the different routes will treat that number differently depending on the numeric accuracy the driver can tolerate, or whether it truncates or rounds up the number to an integer value.

I was able to simulate similar (not identical) weird artifacts like in some of your scans of prints from PS CS6, Paintshop and Photostudio by downsampling Bart's target to 360ppi using 'Nearest Neighbor' resampling in PS. Anyway its pretty obvious from the aliasing pattern in those prints that it was not printed at 720ppi.

I think Roy has confirmed before on forums that QTR does not resize the image data at all, so the operating system likely takes care of that - which may explain to some extent why it looks so different from the others. Looks like bilinear resampling?

Lightroom looks weird, but it does resample before handing the data over to the printer driver, and uses a slightly better downsampling algorithm than Photoshop. Strange artifacts! If not for the smooth gray background it almost looks like a paper feed issue and printhead alignment problem in the star pattern, but since the reproduction of the smooth gray area looks perfect (the scan is too small to tell properly), I assume it is neither, and the other prints collude to agree that the printer is well aligned.

I have sample prints from an Epson 7900 printed from the driver out of PS CS6 of Bart's original camera resolution target, which I know is printed properly at 720ppi and I can safely say it should not look like any of the scans you are showing. The printer can resolve almost as well as my Canon (better in some respects), and the targets look identical when both printers are well aligned.

[samueljohnchia]

Canon iPF is better at resolving fine details, so I'm afraid aligning won't help in this regard.

You may have misunderstood what I said? I agree that my Canon is better at resolving fine details (I don't own any Epson printer), but I can clearly see less than perfect alignment in my prints of Bart's target, and older prints of his camera resolution target show better alignment too, so I think I can do an even better job in the alignment of my Canon.

[samueljohnchia]

The patterns you correctly point at are merely limitations of pixel size, which aren't small enough to depict gradients accurately. Having said that, I am working on generating even higher quality patterns, but they take a lot longer to calculate (processing time x 100) and then need to be composited into the various composite patterns. May take a while to complete.

Not mistakenly, they are supposed to be of equal density (especially the bottom two that run hor/ver), but gamma and profiling may play a trick on us, as well as the difficulty to exactly get intermediate tones from a limited number of inks through dithering. I do see the issue with aliasing patterns, the low frequency waves, and am looking for a way to solve that without compromising the detection sensitivity of errors. Maybe I need to change the overall gamma of the target, or use a different pattern than pure sinusoids (but that would probably need different settings for each dithering method). At the moment I'm calculating the patterns with much higher accuracy to see if that makes a difference. If not, more drastic changes may be needed.

Unfortunately, uniform contrast lines only allow fixed pixel resolution measurements, like 600 PPI = 11.81 cy/mm, 300 PPI = 5.91 cy/mm, 200 PPI = 3.94 cy/mm, 150 PPI = 2.95 cy/mm , 120 PPI = 2.36 cy/mm, 100 PPI = 1.97 cy/mm, and only in horizontal/vertical orientation. But those resolutions should be easy for any printer. It's mainly in the region between 300 or 360 PPi and 600 or 720 PPI that we will see the differences in e.g. media/ink diffusion, and there would be no info there without sinusoidal or other intermediate tones.

I think even in the most well aligned printers the hor/ver lines will print with different densities than the top two diagonal quadrants? But the hor/ver should print with the same density, and the two diagonals should print with the same density assuming all the stars align (bad pun)!

I'm still thinking about the usefulness of the sinusoidal patterns. The low frequency aliasing pattern make them difficult to read at first. Uniform contrast lines are extremely useful to determine issues with dot gain, ink bleed, dot precision etc., because you know they are supposed to be sharp lines. I think the rightmost column could be modified to include them, instead of the diagonal pattern. Those resolutions are very revealing for printers, don't overestimate them! In fact if not for the tiny pattern of hor/ver lines in the top left of your target, I would not have so conclusively determined I could align my print heads better. I also suspect that these inkjet printers have issues firing accurately in a pass depending on what colors/densities are printed in a pass sequence. Nero in another thread on printer linearization made some sort of discovery that indicated something like that I think.

I'm not sure I'm so concerned about detail between 300 and 600ppi or 360 and 720ppi. One has to resample to either size anyway - I want to know how accurately pixels at the highest resolution are rendered. I know my Canon can just about resolve a single pixel line at 600ppi on most gloss or matte media, and any halos will be visible under close scrutiny. I can even observe the mazing/zipper artifacts so common in raw files processed by Lightroom/Camera Raw!

But I am not saying the sinusoids are not useful! After better understanding how they work thanks to your explanations, I can now 'read' print head alignment issues separately from aliasing issues. Brilliant work Bart! This target is a Godsend YOU are a Godsend!

Good. Frankly, your profiling seems to be quite good.

Thank you! After spending many months refining and improving my profiling, I'm glad it has proved fruitful and earned your rare high praise   I think top-notch profiles are still extremely difficult to make even with today's tools, but in general it has become somewhat easier, if one knows what to do. The commercial profiling packages are marketed to make them seem push-button easy to use, but are in fact more complicated than they appear. How is your profiling process coming along? Are you pleased with the measurements of your targets?

I assume that the head travel speed is chosen to minimize any speed splatter on common media. It also ties into the dying/blending characteristics for multiple passes, so timing is probably carefully selected.

It's odd indeed. A long shot, but maybe a print from another application can make a difference? Could there be output dithering taking place by the Color Management system?

I printed using the iPF Photoshop 16-bit Plug-in, which claims to ingest 16 bit data. If memory serves it dumbs it down to 12 bits before handing the data over to the printer - the printer's processor is only capable of dealing with 12 bits of precision. I'm not sure if Canon has updated the LCOA processor to work with 16 bits of data in the latest models. Anyway Photoshop does not work in all 16 bits of precision to start with. This info is several years old from Scott Martin. I am not entirely sure what is going on, but will continue to experiment and maybe get to the bottom of this fuzzy pattern issue. I suspect that text prints better with fewer number of passes. The Highest (max. no. of passes) is more like a special photo mode for the smoothest dither and most uniform dot coverage, and sharp photographic detail. It may not be optimized for fine text and lines. I note that there is a checkbox in the plug-in that is called 'high precision text and fine lines' which may make a difference. I'll look again at my old test prints using this setting.

[Some Guy]

I just downloaded your scan jpegs, they are quite small but still very revealing. I do not think it is a head alignment issue at all.

First up Bart's target for Epson (360/720ppi) printers is not 720ppi - it is 719.988ppi. You need to resize without resampling to 720ppi for one.

If you failed to catch that, I suspect the different routes will treat that number differently depending on the numeric accuracy the driver can tolerate, or whether it truncates or rounds up the number to an integer value.

I was able to simulate similar (not identical) weird artifacts like in some of your scans of prints from PS CS6, Paintshop and Photostudio by downsampling Bart's target to 360ppi using 'Nearest Neighbor' resampling in PS. Anyway its pretty obvious from the aliasing pattern in those prints that it was not printed at 720ppi.

I think Roy has confirmed before on forums that QTR does not resize the image data at all, so the operating system likely takes care of that - which may explain to some extent why it looks so different from the others. Looks like bilinear resampling?

Lightroom looks weird, but it does resample before handing the data over to the printer driver, and uses a slightly better downsampling algorithm than Photoshop. Strange artifacts! If not for the smooth gray background it almost looks like a paper feed issue and printhead alignment problem in the star pattern, but since the reproduction of the smooth gray area looks perfect (the scan is too small to tell properly), I assume it is neither, and the other prints collude to agree that the printer is well aligned.

I have sample prints from an Epson 7900 printed from the driver out of PS CS6 of Bart's original camera resolution target, which I know is printed properly at 720ppi and I can safely say it should not look like any of the scans you are showing. The printer can resolve almost as well as my Canon (better in some respects), and the targets look identical when both printers are well aligned.

This is interesting.  I suspect you are right in that the math involved in the editing software is having an effect, especially in the 14.2 square area where it goes to full black in Zoner software to B&W in PS since both use the Epson driver.  QTR also does some odd effects in that area as well even though it does not use the Epson driver.  Hard to see in the Lightroom that the gray has a reddish tinge to it even though the ICC was off in the Epson driver.  Why the reddish tinge escapes me when it doesn't show in the others.  LR handles this very differently than PS even though they use the Epson driver.

Qimage has the smoothest tonality so it may handle the math better than the others.  I did have the Sharpness turned to "Off" since it uses about a dozen different sharpening methods in the menu.   The "Help" in it says "Vector Sharpening" may be the sharpest, but I didn't go that far.  It has other sharpening methods like Hybrid, Fusion, etc. as well and they all might produce a different look in the end.  I'll probably stay with it as it seems to work the best for me.  The others are very odd in the outcome with the 14.2 square as well as the diagonals in the 8 area on the right with some.  CS6 seemed to produce the finest line detail, but the 14.2 was too disturbing as well as the 8 diagonals.

Interesting test though, but somewhat scary too since the software can really ruin an image as with the B&W squares in 14.2 from the heavy-editing softwares show.  What effect this has does during a normal printing is unknown.

I'm sort of suspicious of the PNG format as it does some compression that may be playing into this when it uncompresses by the sundry software.  PNG may be an act of compromise and leading to the different outcomes?  Maybe an uncompressed TIFF would be better?  I know I had to save the PNG file as a TIFF in Gamma 2.2 for the K7/QTR image which appears better than the other editors images which were all PNG.

I might take one to my pro lab and see what mess they generate.  They use a Mac with a RIP and I use Windows.  Don't know if the math is the same between them and the final outcome.

SG

[Bart_van_der_Wolf]

First up Bart's target for Epson (360/720ppi) printers is not 720ppi - it is 719.988ppi. You need to resize without resampling to 720ppi for one.

Hi Samuel,

I missed that, sorry, because I specifically changed the PPI tag to 720 before saving and assumed it was as instructed. I've done that again, checked that the tag now says 720, and uploaded the image again for future downloads.

And checking it again, the tag was changed once more, presumably because Photoshop doesn't know how to deal with 130mm, 3685 pixels, and then calculates the PPI from there (which then becomes 719.9923077 PPI, and not what Photoshop makes of it), instead of accepting the user's input for the PPI tag field. When is an international company like Adobe going to adopt the worlds standard units of measurement, or is it a TIFF conventions thing (also Adobe owned)? PNGs should just do what they are told, but the PPI tag is already changed by Photoshop.

Anyway, a decent printing application should ignore the tag anyway, and use the output size as specified by the user, or 100% of the pixel dimensions, or use an enforced PPI setting.

Interesting what such a test target can reveal.

Cheers,
Bart

[Some Guy]

Bart, do you have this file as a TIFF?  As I read it, the PNG (even PN-24) is an 8 bit and the TIFF is larger 16 bit.

Perhaps the 8 bit math is doing something within the software resulting in the sundry software outcomes against a 16 bit file?  I don't know why the PNG > TIFF conversion seemed to work better in K7/QTR.  If QTR did PNG, no doubt it would be a different result than the converted to TIFF one.

SG

[Bart_van_der_Wolf]

I think even in the most well aligned printers the hor/ver lines will print with different densities than the top two diagonal quadrants? But the hor/ver should print with the same density, and the two diagonals should print with the same density assuming all the stars align (bad pun)!

I agree, not about the pun but, what is most important to look for is asymmetry, like between the horizontal and vertical cycle areas, or between the two diagonal cycle patterns. I am at the moment busy (between regular work) with improving the sinusoidal patterns even more, because I think they can still add a lot of insight, but I want them to be as flawless as possible and certainly not the weakest link. The recalculation of the patterns now takes much(!) longer though, so I only get output presented by the computer (cooling fans are working hard, and all CPU cores are kept busy) for further editing/compositing every once in a while. Hopefully I can also reduce the difference between diagonal and horizontal/vertical cycles, so they don't distract as much.

I'm still thinking about the usefulness of the sinusoidal patterns. The low frequency aliasing pattern make them difficult to read at first. Uniform contrast lines are extremely useful to determine issues with dot gain, ink bleed, dot precision etc., because you know they are supposed to be sharp lines. I think the rightmost column could be modified to include them, instead of the diagonal pattern. Those resolutions are very revealing for printers, don't overestimate them! In fact if not for the tiny pattern of hor/ver lines in the top left of your target, I would not have so conclusively determined I could align my print heads better. I also suspect that these inkjet printers have issues firing accurately in a pass depending on what colors/densities are printed in a pass sequence. Nero in another thread on printer linearization made some sort of discovery that indicated something like that I think.

I'll have to see where it takes us, and if the aliasing can be traced to what's causing it. The sinusoids have several very interesting properties, and some of those have not been addressed yet. It has to do with upsampling and how the top left pattern (11.8 or 14.2 cy/mm depending on the PPI version) gets rendered. The higher resolution diagonals may look a bit funky at close inspection of the original on display, but they turn into (almost) perfect sinusoids when properly upsampled.

I'm not sure I'm so concerned about detail between 300 and 600ppi or 360 and 720ppi. One has to resample to either size anyway - I want to know how accurately pixels at the highest resolution are rendered.

I understand, but to judge resampling we need a predictable pattern that needs to look exactly the same as the original, only at a different size. If the pattern is not predictable, or not as 'organic' as a regular image but has only straight lines, then the challenge is a rather different one, and maybe easier (when less intermediate color blending/dithering is required).

But I am not saying the sinusoids are not useful! After better understanding how they work thanks to your explanations, I can now 'read' print head alignment issues separately from aliasing issues. Brilliant work Bart! This target is a Godsend YOU are a Godsend!

Blush... Well, I try to make a small contribution, regardless of who sent me or not.

Thank you! After spending many months refining and improving my profiling, I'm glad it has proved fruitful and earned your rare high praise   I think top-notch profiles are still extremely difficult to make even with today's tools, but in general it has become somewhat easier, if one knows what to do. The commercial profiling packages are marketed to make them seem push-button easy to use, but are in fact more complicated than they appear. How is your profiling process coming along? Are you pleased with the measurements of your targets?

Slowly getting there. The prints are dry, but I haven't had the time to let i1Profiler do its job yet. Another printer is running low on ink, so I want to preserve that for actual image prints for the moment (note to self, need to order some new ink as well).
And then there are comparisons needed between different printing applications, Qimage and Photoshop/Lightroom for example, to see if and what is different.

I printed using the iPF Photoshop 16-bit Plug-in, which claims to ingest 16 bit data.

I understood that, but maybe the application that sends it data to the printer driver/plugin already does something we do not want.

Cheers,
Bart