Up-Rezing or Down-Rezing an image

[disneytoy]

Hello!

This is fascinating. I am working on printing some 16mp images on a 9890 to40x60". I'm trying to figure the best workflow to getting the best blow up.

I hadn't thought about targeting the output resolution i.e 360dpi. (720dpi would be impossible)

That being said. If I upres my images to the final print size @360dpi, is there any benefit "when?" the upsize occurs?

Example.

These images have had a fair amount of processing. Lightroom adjustments, Upright, Noise reduction, Clarity, Sharpening, cloning, spot healing, some external plug-ins like NIK Define. I am even playing with adding real grain to give a realistic grain pattern to the 40x60"

So, I have been upsizing via the latest PS CC to the print dimension @240dpi. A far stretch. But that is the size I need for the show.

I'm wondering if upresing would be better done before one of the heavy images processes, like clarity, or noise reduction? I think Sharpening would be best at the end.

Thanks to you guys I will be much more careful in targeting my final DPI to 360.

In fact I think I may have output my final Tiff files @240, but ended up in Lightroom trying to print them with a smaller border (minor enlarging), essentially, screwing up the 240 dpi output.

You can learn a lot on this forum.

Thanks

Max

[dwswager]

Sadly, this only perpetuates the myth. We know better now. It is better to send 360PPI than any other multiple of 60. That's the old myth. And yes, it IS better to upsample to 360 (or 720 using finest detail) than letting the driver deal with non-standard rez.

Sorry, but it's my experience the military doesn't always get the correct info about much of anything that doesn't actually shoot.

:~)

I take no offense, I was just relaying what was divulged during a program kickoff meeting.

I do ask where the definitive information that it is better to up sample to 360ppi or 720ppi originates?  Is this something divulged by Epson publicly or privately based on knowledge of the dithering algorithm or testing or does this come from your considerable experience?

Thanks!

[hugowolf]

... I am working on printing some 16mp images on a 9890 to40x60". I'm trying to figure the best workflow to getting the best blow up.

...

Generally: uprez and then perform output sharpening just prior to printing, has been the perceived best practice. It is the way the Lr and Qimage do it for you. There are several reasons. If you do a search, you find it was discussed a lot.

[ppi, not dpi]

Brian A

[Schewe]

I take no offense, I was just relaying what was divulged during a program kickoff meeting.

I do ask where the definitive information that it is better to up sample to 360ppi or 720ppi originates?

Well, a lot of the discussion occurred here on LuLa and that lead me to do the research to write an article in Digital PhotoPro magazine called The Right Resolution.

[Bart_van_der_Wolf]

I do ask where the definitive information that it is better to up sample to 360ppi or 720ppi originates?

Extensive testing, and the fact that the printer driver tells the application that is using the printer driver what resolution it expects to be sent.
Sending a regular pixel pattern to the printer will result in resampling/aliasing artifacts, unless exactly 360 or 720 PPI is used on Epsons, 300 or 600 PPI on Canon/HP and others. The principle of native printer resolution has been known for a long time, and has e.g. been used by Qimage since 1998.

Otherwise the printer driver (supported by the OS on Mac) resamples to 360 PPI (or 720PPI when finest detail is selected). This is a separate issue from the 60 dot dither pattern that is used to create intermediate ink colors. The resampling method that the printer driver uses is relatively simple (usually bi-linear), and it doesn't allow to output sharpen after resampling, so it's best to resample first, then sharpen, then send to the printer.

Cheers,
Bart

[dwswager]

Extensive testing, and the fact that the printer driver tells the application that is using the printer driver what resolution it expects to be sent.
Sending a regular pixel pattern to the printer will result in resampling/aliasing artifacts, unless exactly 360 or 720 PPI is used on Epsons, 300 or 600 PPI on Canon/HP and others. The principle of native printer resolution has been known for a long time, and has e.g. been used by Qimage since 1998.

Otherwise the printer driver (supported by the OS on Mac) resamples to 360 PPI (or 720PPI when finest detail is selected). This is a separate issue from the 60 dot dither pattern that is used to create intermediate ink colors. The resampling method that the printer driver uses is relatively simple (usually bi-linear), and it doesn't allow to output sharpen after resampling, so it's best to resample first, then sharpen, then send to the printer.

Cheers,
Bart

So it is empirically derived and not definitive.  Fine. I certainly respect Jeff and your experience and results!

But the massive assumption being made, with no reference to back it up, is that the printer driver re-samples the original data to something whether 360, 720 or something else which can't be learned empirically!

[Bart_van_der_Wolf]

So it is empirically derived and not definitive.  Fine. I certainly respect Jeff and your experience and results!

But the massive assumption being made, with no reference to back it up, is that the printer driver re-samples the original data to something whether 360, 720 or something else which can't be learned empirically!

There are ways to learn it, e.g. this user found out that 359 PPI and 361 PPI get resampling artifacts, 360 PPI doesn't. He also mentions a comment by Eric Chan (member of the Lightroom development team) about resampling to 360 PPI,  and here some more examples from printing at 240 PPI compared to software upsampled 360 PPI. There are more threads and examples, but I don't have the time to collect them for you.

As said, printing a line or dot grid will tell if resampling is happening, and if head alignment and paper transport is well adjusted one can derive from the interference pattern what has happened. You can already simulate that by resizing such a grid in e.g. Photoshop with bilinear resampling, and compare it to what the printer produces.

Cheers,
Bart

P.S. I've attached a resampling test image that should produce aliasing in proportion to the number of pixels resize difference.
P.P.S. A zoneplate target will allow to detect all sorts of aliasing issues at various angles, but it may be harder to pinpoint the exact factors involved due to printers using (stochastic) dithering patterns.

[Ernst Dinkla]

For the HP Z3100 driver I tested it in downsampling as discussed here:
http://forum.luminous-landscape.com/index.php?topic=54798.60
Alternating lines were lost of the 1200 PPI test target due to aliasing in the conversion to 600 PPI, the same for a 600 PPI test target  to 300 PPI. The route from the application to the printer starts with a straight downsampling, most likely nearest neighbour before the dithering etc builds the print data. The difference with upsampling is better algorithms used for the last in today's drivers.
Printing from Qimage with its resampling and sharpening off. I doubt I still have the prints here but could check.

It is as empirical as any test done to check a theory.

Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
December 2014 update, 700+ inkjet media white spectral plots

[Bart_van_der_Wolf]

It is as empirical as any test done to check a theory.

Indeed, it's a bit like gravitation. Nobody tells you what is really is, but the empirical evidence and some clever deduction produces a workable theory (proposed by Einstein) that even allows to navigate in outer space.

Cheers,
Bart

[dwswager]

There are ways to learn it, e.g. this user found out that 359 PPI and 361 PPI get resampling artifacts, 360 PPI doesn't. He also mentions a comment by Eric Chan (member of the Lightroom development team) about resampling to 360 PPI,  and here some more examples from printing at 240 PPI compared to software upsampled 360 PPI. There are more threads and examples, but I don't have the time to collect them for you.

As said, printing a line or dot grid will tell if resampling is happening, and if head alignment and paper transport is well adjusted one can derive from the interference pattern what has happened. You can already simulate that by resizing such a grid in e.g. Photoshop with bilinear resampling, and compare it to what the printer produces.

Cheers,
Bart

That is what empirical means...by observation or experience.  You try different things and look at the output and choose the input that gives the most favorable output.  We guide hyper-velocity vehicles not by known math, but by empirical results built up over decades of testing.

But there is a big difference between saying upsizing to 360 or 720ppi produces the best results based on mountains of testing and the printer driver resamples to 360 or 720ppi.   The first is true based on observation, but the second can not be determined true by observation.  It is a theory that might be true. The reason I point this out is that it is actually unnecessary for a dithering algorithm to resample to some base to calculate the dithering pattern.  We do it by matching input and output area and converting directly. 

Having started with the 1994 Seiko Epson Stylus Color printer and currently still stuck on an R2400, I have always fed my printer 360ppi when printing for maximum quality.  I will upsample to 360ppi and even down sample to it.  Depending on the original image, I find that upsampling say a 430ppi image to 720ppi does not provide increased benefit than down sampling it well to 360ppi.  Though I might upsample to to 720 if stating closer to 500ppi.

Assuming a 10% drop size spread, then to print at 720 would require drops on paper of 35.28 microns each which relates to a 32.07micron drop diameter.  Calculating out the volume of a droplet would be about 17picoliter.  Hence, the 3.5picoliter dropsize allows for printing at 720dpi. Conversely, if I did the quick and dirty math correctly, then 3.5ml drops would equate to roughly 19 microns in diameter would allow printing at roughtly 1350dpi.  Of course, the paper has to be able to take that kind of ink load and the printer have that type of horizontal and vertical mechanical precision.

[Ernst Dinkla]

I see the R2400 was introduced around 2005. In docs Epson quotes resp. 3 and 3.5 picoliter for the dropsize, that is the minimum droplet size, usually 2 other sizes above that size. In the highest print resolution setting only the minimum droplet size is used. On the old Epson 9000 the print resolution was documented as 1440 dpi, in practice the printer laid down 720 dpi, the next row shifted half the droplet pitch and another 720 dpi was laid down. Whether that could be qualified as 1440 dpi was questionable. The minimum droplet size was about 10 picoliter, variable droplet sizes still in the early stages of development. Whether droplets of one ink channel are allowed to overlap or not is also open for debate. In the past the ink amount laid down with different print resolutions required different custom profiles. Today for several models this is compensated it seems. either in dithering or media preset ink limits. So what you might conclude of your empirical testing with the R2400 may not work for general conclusions on the behavior of today's printer/driver models.

In the thread I gave the URL for you can read that Jeff Schewe even found a 1440 PPI input printer quality setting in Epson x900 models for the Epson Proofing White SemiMatte, a very high quality paper. Most likely the driver reporting to Lightroom.  http://forum.luminous-landscape.com/index.php?topic=54798.180 The Epson engineers probably thought it was worth it with that quality of paper. I found an 1200 ppi input resolution in an older HP Z3100 driver but it was gone with the next driver version installed. Both at the edge of what can be gained in image quality I guess.

My downscale aliasing test was done to check whether the theory of fixed input resolutions was sound. A theory developed as no hard information from manufacturers reaches the users. A theory that forms the base of the extrapolation functions Mike Chaney added to Qimage since more than a decade. So here the theory existed and I made the test to check it. So I wrote "as empirical" lifting that proof somewhat above the empirical evidence of print quality observations.

What is interesting is the QTR driver based on Gutenprint/Gimpprint. When asked Roy Harrington wrote it did not have that kind of input resolution steps. I can imagine that guys like Robert Krawitz used another approach for that print engine. We do not argue that a driver has to have certain input resolution steps but find evidence that OEM drivers have them and that they force the data through less than ideal resampling algorithms before the real driver job begins. Edit: well I checked my memory of this and found a reply by Roy that QTR resamples everything thrown at it to 720x720 before it goes into the dithering phase for the printer resolutions set. An avant la lettre and hard coded equivalent for the Bart van der Wolf / Jeff Schewe approach.

Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
December 2014 update, 700+ inkjet media white spectral plots

[Bart_van_der_Wolf]

But there is a big difference between saying upsizing to 360 or 720ppi produces the best results based on mountains of testing and the printer driver resamples to 360 or 720ppi.   The first is true based on observation, but the second can not be determined true by observation.

Hi,

In the way that we have iron clad proof (e.g. Epson documentation from the printer driver programmers), no, you are right. However, there's that thing popularly called the duck test. People like Jim Kasson whom's blog I linked to, even dared to call his duck 'nearest neighbor' resampling, because that's the closest resemblance to what the printer he tested produced. Maybe in the mean time that has been improved to bi-linear, but that's not the relevant point.

Not only can we do much better resampling ourselves, but we are using inductive reasoning to qualify the method used, without hard proof, just duck proof, highly probable. The conclusion that resampling is used, is consistent with the aliasing artifacts formed by that method. Hence we call the duck, 'resampling'.

It is a theory that might be true. The reason I point this out is that it is actually unnecessary for a dithering algorithm to resample to some base to calculate the dithering pattern.  We do it by matching input and output area and converting directly.

Sure, many methods can be used, but given the 'quality' of the output, it's unlikely that they used something more sophisticated. Besides, it's easier and faster to present the dithering algorithm a known quality (maximum resolution 180/360 cycles per inch, or 360/720PPI, potentially 41 percent higher diagonally) of data, to allow for the seamless successive printhead passes. Complexity usually results in added (maintenance and buffer memory) cost and slower performance, which are things that are not popular in production environments.

Cheers,
Bart

[dwswager]

Not only do I not doubt Jeff and Bart's recommendations, I've been using them since before I knew who Jeff and Bart were.

My comments only relate to what we know versus what we think.  There is no need to say that the printer resamples to X to make the point that sending the printer Xppi results in the best output.  That has been demonstrated empirically in response to a lack of definitive answers from Epson.  I would suspect Epson was leery of saying too much in 1994 when no one really had much high density data.  Let's face it, if people are shooting a 2.1MP $7000 Nikon D1 do you really want to tell they they are woefully short on data to use the printer at it's most effective.  But along the way, I think it has been irresponsible of Epson and other manufacturers not to specifically document how to get the best output from their printers.

I see the R2400 was introduced around 2005. In docs Epson quotes resp. 3 and 3.5 picoliter for the dropsize, that is the minimum droplet size, usually 2 other sizes above that size. In the highest print resolution setting only the minimum droplet size is used. On the old Epson 9000 the print resolution was documented as 1440 dpi, in practice the printer laid down 720 dpi, the next row shifted half the droplet pitch and another 720 dpi was laid down. Whether that could be qualified as 1440 dpi was questionable. The minimum droplet size was about 10 picoliter, variable droplet sizes still in the early stages of development. Whether droplets of one ink channel are allowed to overlap or not is also open for debate. In the past the ink amount laid down with different print resolutions required different custom profiles. Today for several models this is compensated it seems. either in dithering or media preset ink limits. So what you might conclude of your empirical testing with the R2400 may not work for general conclusions on the behavior of today's printer/driver models.

In the thread I gave the URL for you can read that Jeff Schewe even found a 1440 PPI input printer quality setting in Epson x900 models for the Epson Proofing White SemiMatte, a very high quality paper. Most likely the driver reporting to Lightroom.  http://forum.luminous-landscape.com/index.php?topic=54798.180 The Epson engineers probably thought it was worth it with that quality of paper. I found an 1200 ppi input resolution in an older HP Z3100 driver but it was gone with the next driver version installed. Both at the edge of what can be gained in image quality I guess.

My downscale aliasing test was done to check whether the theory of fixed input resolutions was sound. A theory developed as no hard information from manufacturers reaches the users. A theory that forms the base of the extrapolation functions Mike Chaney added to Qimage since more than a decade. So here the theory existed and I made the test to check it. So I wrote "as empirical" lifting that proof somewhat above the empirical evidence of print quality observations.

What is interesting is the QTR driver based on Gutenprint/Gimpprint. When asked Roy Harrington wrote it did not have that kind of input resolution steps. I can imagine that guys like Robert Krawitz used another approach for that print engine. We do not argue that a driver has to have certain input resolution steps but find evidence that OEM drivers have them and that they force the data through less than ideal resampling algorithms before the real driver job begins. Edit: well I checked my memory of this and found a reply by Roy that QTR resamples everything thrown at it to 720x720 before it goes into the dithering phase for the printer resolutions set. An avant la lettre and hard coded equivalent for the Bart van der Wolf / Jeff Schewe approach.

Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
December 2014 update, 700+ inkjet media white spectral plots

[Schewe]

Jeff Schewe even found a 1440 PPI input printer quality setting in Epson x900 models for the Epson Proofing White SemiMatte, a very high quality paper.

Actually, a fellow from Epson told me that they added that specific settings for proofing papers for the express purpose of being able to print actual halftone dots. And I tested it and could not find any benefit for printing regular photos at that setting. The reason I tested it was to determine of LR's max rez should go from 720 to 1440 ppi.

[bjanes]

Well, a lot of the discussion occurred here on LuLa and that lead me to do the research to write an article in Digital PhotoPro magazine called The Right Resolution.

Jeff,

That link appears to be broken.

Regards,

Bill

[Schewe]

That link appears to be broken.

Hum...well go to http://www.digitalphotopro.com click on the Workflow link under Technique tab and scroll down to The Right Resolution article.

Maybe I miscopied the url when I added the url link...Not sure but DPP may be doing something with their urls, but http://www.digitalphotopro.com/technique/photography-workflow/the-right-resolution works for me.

[dwswager]

Hum...well go to http://www.digitalphotopro.com click on the Workflow link under Technique tab and scroll down to The Right Resolution article.

Maybe I miscopied the url when I added the url link...Not sure but DPP may be doing something with their urls, but http://www.digitalphotopro.com/technique/photography-workflow/the-right-resolution works for me.

With respect to the linked article, you wrote: The Epson pro printers' print heads have a reported output resolution of 360/720 dpi, depending on the print mode.

Where has this been reported?  I fear this is one of those statements that someone says and it gets repeated over and over until the conjecture is believed to be fact.  In over 20 years of using Epson Stylus printers and working directly with Epson, I have never heard them make this claim.  What they do claim is that the printers have the ability to print X number of different droplets per linear inch depending on the printer and printer settings.  Epson printers print with variable drop volumes with variable spacing.

Later, the following statement appears: The people at Epson say that the print driver doesn't do the re-sampling, and since the application sending the print doesn't do the resampling unless asked to do so, the resampling must be happening in the print pipeline.

This is an assumption that the image is resampled which may or may not occur.  But I assure you it is not the Windows GDI doing it if it does occur.  Either the printer driver does it or it doesn't occur.  In fact, on Windows, Epson installs it's own printer port and not only bypasses GDI, but also includes it's own spooler.  

Considering that you write that Epson denies the the driver resamples the data and we know that a dithering algorithm does not require the data be resampled to a specific base resolution, is it more likely that the data isn't resampled or some other mystery cuplrit in every operating system just happens to resample to a somewhat bizzare 360/720ppi?  Otherwise, print quality would be different on every OS!  And if Both major OS's resampled the data to 360/720 why would Canon, HP and every other printer manufacturer design their printers to use 300ppi where now they would have to resample the data back down from the previous re sampling.  [ADDED]  Of course, the other alternative is that Epson lied and the driver does re-sample!

All that said, I generally send 360ppi data to my Epson printers because I believe (but do not know) it provides the best quality.

[Schewe]

With respect to the linked article, you wrote: The Epson pro printers' print heads have a reported output resolution of 360/720 dpi, depending on the print mode.
Where has this been reported?

It's reported to the system...it's well known and documented. You can prove it to yourself by saving images as PDFs from the Mac print dlog and then checking the resulting resolution. If you don't believe me, Mike Chaney has written about this...while Epson doesn't include this in written docs, it ain't rocket science to prove...

This is an assumption that the image is resampled which may or may not occur.

This is exactly what an Epson product manager said to me when asked is the Epson driver "resampled" and image that wasn't sent as 360/720 PPI.

Clearly an image sent as anything other than 360/720 to an Epson Pro printer is modified and can result in some image artifacts (particularly stair-stepping). Whether or not the image is resampled has not been confirmed by either Apple nor MSFT...Epson says no...and I know for a fact neither Photoshop nor Lightroom does. So if not the driver nor the app, where is the "resampling" occur?

Wherever it occurs, it's clear that the resampling (if it occurs) is substandard. Some say Bilinear while other say Nearest neighbor. Which ever, it tends to suck...

I'm still of the mind that the driver doesn't "resample" but merely sends the the raw image data to a complex dithering algorithm that tends to add artifacts if the data isn't reported as either 360 PPI (or 720 PPI if finest detail is selected).

Also note that Epson printers that do not have finest details as a driver option (the Epson consumer printers) can't report 720 and thus can't benefit from resolutions beyond 360 PPI.

[Bart_van_der_Wolf]

With respect to the linked article, you wrote: The Epson pro printers' print heads have a reported output resolution of 360/720 dpi, depending on the print mode.

Where has this been reported?

It is reported by the printer driver to the program that interrogates the driver. The PPI that's being fed back, depends on the driver settings, e.g. paper/ink combinations and resolution settings (Epson's 'Finest Detail' option to allow 720 PPI', Canon usually specifies in the driver what's used, e.g. 600PPI).

Later, the following statement appears: The people at Epson say that the print driver doesn't do the re-sampling, and since the application sending the print doesn't do the resampling unless asked to do so, the resampling must be happening in the print pipeline.

This is an assumption that the image is resampled which may or may not occur.  But I assure you it is not the Windows GDI doing it if it does occur.  Either the printer driver does it or it doesn't occur.  In fact, on Windows, Epson installs it's own printer port and not only bypasses GDI, but also includes it's own spooler.

Things differ between Windows and Mac platforms. On Macs part of the print processing is delegated to OS related functions that are used for resampling (see this thread), while on Windows it's typically a printer driver operation. On Windows it is possible to use a 16-bit/channel print output (e.g. Canon supports that with their XPS drivers), I believe on Macs the maximum is currently 8-b/ch. So there are differences.

Cheers,
Bart

[mchaney]

I've been asked to come here and chime in.  I see that the thread has branched out in many different directions so I'm not sure exactly which aspect to address, so I'll address the original question.  The original question, as I see it, was regarding whether or not it ever makes sense to downsample to something like 120 PPI for something like an Epson printer.  Downsampling to 120 PPI means that you started with something higher than that and you are downsampling to 120 PPI.  As an example, you start with an image that happens to be 217 PPI at the intended print size.  So is there any benefit to downsampling from 217 PPI to 120 PPI versus upsampling from 217 PPI to 360 or 720?

I would say no.  The reason being, whenever you downsample, you are throwing away data.  That's data you cannot get back and at something as low as 120 PPI, the "lost data" will be evident in the print if you compare to one that was upsampled instead to say, 360 or 720 PPI.  The more data you start with, the more "detail" the viewer will perceive (up to a point of course, depending on viewing conditions, device limitations, etc.).  On the other side of the fence, you would choose to upsample to 360 or 720 not because you think you are "gaining data", but because those resolutions align nicely with the dot patterns used by an Epson printer and, being multiples, align well with those dot patterns and produce fewer artifacts.  As an added benefit, you are using the data you have and not throwing anything away when you upsample.

When talking about upsampling to 360 versus 720, I've found that most printers/papers can render detail at a maximum of somewhere between that range.  In other words, some detail can be rendered higher than 360 but probably not as high as 720.  It's not straightforward because it depends on the printer you are using, the paper you are printing on, the driver settings, and even the color you are testing.  Generally speaking, if you are printing a high res image and that image already has higher than 360 PPI at the intended print size, printing at 720 versus 360 will yield a little better minute detail.  But you'll probably have to put your nose almost to the paper to see it.  That's why I often say that printing at 360 PPI is fine for large prints and 720 more appropriate for smaller ones because those smaller ones might undergo closer scrutiny.  Nothing wrong with printing at 720 PPI for everything, but you may never see the benefit on a 40x30 hanging on a wall.

Mike