Upsampling and output sharpening

[Arlen Thomason]

I've been running some tests with PK Sharpener on larger prints, and would like some feedback as to whether I'm on the right track.

I want to print some Canon 10D images at 11x16.5". The native 10D file resolution for this size is 186.182 ppi. I have tried 3 procedures for generating a file that I then print via ImagePrint 6 on an Epson 2200 and Epson Professional Glossy paper. In each case, the starting RAW image was converted in ACR, brought into PS CS at native resolution and 16 bits, capture sharpened, tonally corrected, and then:

1) Resampled down via Bicubic Sharper to 180 ppi, followed by Output Sharpener Inkjet 180 Glossy.

2) Resampled up via Bicubic Smoother to 240 ppi, followed by Output Sharpener Inkjet 240 Glossy.

3) Resampled up via Bicubic Smoother to 360 ppi, followed by Output Sharpener Inkjet 360 Glossy.

In each case, during output sharpening I examined the resized image onscreen at 50%, and adjusted the opacity of the layers to what appeared optimum to my eye, and attempted to make the "sharpness" of each image approximately equal. I then compared the resulting prints.

What I found was that at 180, the default Output Sharpener settings (100% master opacity, both contours 50%) yielded optimal sharpening. At 240, each contour had to be raised to about 80% to yield an equivalently sharp print. At 360, even setting both contours to 100% was insufficient sharpening, and I had to run the 360 sharpener a second time to get it as sharp as the other prints.

Other than the amount of sharpening required, I could see no significant differences in the prints from images sent to IP at 180, 240 or 360 ppi. I am therefore tentatively concluding that for prints at this size, there is no advantage to upsampling, and that sharpening is more straightforward when upsampling is avoided. The default settings of the output sharpeners appear to be geared for images which have not been upsampled.

Any comments on whether my observations & conclusions are consistent with others? If there are advantages to upsampling and sharpening that I am missing, I would appreciate having them pointed out.

[Bruce Fraser]

I wouldn't say that there are never any advantages to upsampling, but it's pretty rare.

Our output sharpeners are indeed optimized for files containing real data, rather than upsampled ones.

However, to print a 10D file at that size, if you stay at native resolution, I wouldn't bother downsampling from 186.182 ppi to 180?the difference is so small that it's not worth either the speed hit or the pixel-munging.

Resampling is something I only do when I absolutely have to. It's like the talking dog?what's noteworthy is that it talks at all, not how well it does so!

[Arlen Thomason]

Thanks Bruce. So with regard to the often mentioned recommended printing image resolutions of 180, 240 and especially 360, one should only change to them if downrezzing, and even then don't bother if you are already close to one of them?

[Bruce Fraser]

Exactly!

[kdoc]

I'm not clear on what you're saying: let's give a few so-called (incorrectly) "scenarios" for printing.

1. You're at 200 after selecting the dimensions for printing: do you uprez to 240, downres to 180 or leave as is?

2. You're at 185: Do you feel that 185 (close to 180) is sufficient res for printing or do you uprez to 240?

3. You're at 285: Leave alone, bring to 240; or up to 300?

In effect: is 180 really enough? And when do you leave alone and when do you uprez or down rez--assuming you have an image dimension you know you want to print at.

Thank you.

kdoc

[Bruce Fraser]

I'm not clear on what you're saying: let's give a few so-called (incorrectly) "scenarios" for printing.

1. You're at 200 after selecting the dimensions for printing: do you uprez to 240, downres to 180 or leave as is?

I'd leave as is and use the inkjet 180 output sharpener.

2. You're at 185: Do you feel that 185 (close to 180) is sufficient res for printing or do you uprez to 240?

I'd leave as is and use the inkjet 180 output sharpener.

3. You're at 285: Leave alone, bring to 240; or up to 300?

I'd leave as is and use the inkjet 300 output sharpener.

In effect: is 180 really enough? And when do you leave alone and when do you uprez or down rez--assuming you have an image dimension you know you want to print at.

If I were printing a 4x6 from a 22MP capture, I'd downrez to 480 ppi. I haven't found that it's ever worth sending more than 480 ppi. If I really need a big print from a small file that contains a lot of high-frequency info, and 180 ppi really isn't cutting it, I might try to uprez to 240, but I basically try to avoid uprezzing.

180 ppi works pretty well for larger prints viewed at normal viewing distances (i.e., not viewed by someone whose nose is touching the print with a loupe screwed into their eyeball). More real data is desirable. Uprezzing and resharpening is a desperate move I only try when all else fails.

Bruce

[kdoc]

Very helpful Bruce: Implicit in what you said, and you might want to elaborate a little on this point, is that the (Epson 2200 or up) printer really doesn't care whether it gets an even multiple of its favored size (such as 180, 240, 300, 360 (which is supposed to be the 2200 "native resolution"), 480--that the printer handles 250 or 260 as well as it handles 240 (except, of course, for the slightly different resolution)--that an exact multiple doesn't matter! Is that correct?

You might also wish to comment (or not to) on why there's some disagreement in this field. For example, Jullienne Kost recommends taking almost any size and up-rezing to 360 when she uses the 2200. Is she just being over-general for teaching purposes, or...? Several people out there uprez with impunity, especially with the advent of the new bicubic smoother and sharper. It sounds like you feel that interpolation is quite destructive and visible.

kdoc

[Bruce Fraser]

If you're printing synthetic targets comprised of BW line pairs, you'll see a benefit to printing at an integral divisor of the printhead resolution. In my experience, that benefit doesn't translate to printing natural photographic imagery.

It's not that uprezzing degrades the image so much as that it's a huge amount of extra work for marginal, if any, benefit. If I see aliasing on the print, I'll uprez and (carefully) resharpen in an attempt to eliminate it. I think I've done that maybe three times in the last five years.

I've no idea why Julianne is making that recommendation. I'll certainly ask her the next time I see her.

[alfin]

A very useful discussion guys, thanks! I think many of us struggle with this type of questions for most of our prints. Bruce, this is the kind of knowledge missing in the PKS help file. Just a small hint...

BR/Lars

[Bruce Fraser]

I'm hesitant to put stuff like this in the manual simply because it's anecdotal evidence based on my own practices rather than the product of rigorous research. Most of the time, I do wind up printing at an integral divisor of the printhead, but maybe 15% of the time I don't. A few quick experiments were enough to convince me that uprezzing to get to an integral divisor wasn't worth the candle. I encourage anyone who cares enough to do the same experiments.

Someday I'd love to have the time to really test this rigorously, but in the meantime I'd prefer to avoid making definitive statements because they inevitably get taken as invitations to pissing matches, and life is too short for those.

[Son]

I am confused regarding the maximum PPI you can send to an inkjet printer. Some writers suggest that 360PPI is native resolution for inkjet printers and sending more pixels only slows the printer down because the printer has to downsample the information internally. Photokit Sharpener has a 480PPI sharpener routine. Is 480PPI the max you should send to an inkjet printer?

[Bruce Fraser]

In the testing we did for the output sharpeners, we saw a small but noticable benefit to sending 480-ppi of REAL DATA (i.e., no upsampling) to the 2880-dpi epsons when printing through the Epson driver. In theory that shouldn't be the case, but reality beats theory every time.

Some people who should know have told me that the 480-ppi performance is a result of a bug in the Epson driver. They may well be right. I've tried sending 480 ppi through ImagePrint, and I don't see the same benefit I do with the Epson driver, so for IP I always use 360ppi or less.

I've seen no benefit to sending more than 480 ppi, ever. I generally use 480 ppi only for fairly small (?8x10) prints, and again, I emphasize, only when I have real pixel data at that resolution.

[sgingold]

Bruce
If I may ask this even though it has sort of been covered in another thread-
If I scan a slide at 4000 ppi and in PSCS set an image size of 7.5 x 10.6, the resolution becomes @467 with no resampling. Am I better off at that juncture output sharpening at 480 to my 2200? It sounds as though if I do that I should be using the 2800 instead of 1400 output in the Epson driver. Some have suggested downsampling to 360, but that would seem to be throwing out detail.
These are landscape images both large subject and macro detail.
Thanks for your thoughts.
Steve Gingold

[Bruce Fraser]

I'd use the inkjet 480 sharpener and print with no resampling. If you're printing on glossy paper, you'll see a tiny benefit to using 2880 over 1440?that's what the InkJet 480 Glossy Fine sharpener is for. Whether it's worth the extra time it takes to print is something only you can decide.

If you aren't printing on glossy paper, I very much doubt whether you'll see a difference between 1440 and 2880?we didn't in fairly exhaustive testing.

[sgingold]

Bruce
Thank you for the quick reply.
I print mostly on Premium Lustre and Enhanced Matte and had been staying with 1440. At 8 x 10 I doubt that there would be a significant difference. Thanks for the advice
Steve Gingold

[Bruce Fraser]

We didn't see any benefit to 2880 on anything except Epson Premium Glossy.

[Mark D]

I am using Epson 4000 and PK Sharpener Pro with Canon 1Ds files. I print on Epson Enhanced Matte at 1440 dpi. I have several observations and a question.

(1) How many PPI to send to the printer? I was concerned about this, but not any more. I was told by an experineced professional who should know (not at Epson - so far they have failed to answer such questions) that the printer throws away anything more than 360 PPI, so if the file has more than this at the intended print size, better to down-res with bicubic sharper in Photoshop because it does a better job than the Epson printer driver. So I tried BOTH: (a) leaving the file at 451 PPI with no downrezzing, PK Output Sharpener at 480; (b) dowrezzing to 360 in Photoshop and PK Output Sharpener at 360. The prints were 6 * 9 inches image size. Looking at these results with no loupe, there was no detectable difference. Using a 10X loupe there was a small positive quality difference in the mosre minute details in the 360 print. So perhaps this means that for finer detail on big enlargements the advice I got makes some sense.

(2) I have achieved commercial quality results from the Epson 4000 sending ANY PPI SPECIFICATION in the range of 180 to 480 to this printer from A3 size downward. Much depends on the image content also.

(3) PK Sharpener Pro Output Sharpener seems to break down completely if the print size is very small (e.g. 4*6 inches) and the resolution very high (e.g. above 550-600PPI). Using the highest setting 480, the result on the monitor is tremendous speckling and very brittle imagery. I know the monitor is not a reliable indicator of the impact of sharpening, but this was so extreme it would obviously be a waste of paper printing it. I therefore assume that for such cases one MUST downrez to use Output Sharpener successfully. Is this a fair conclusion?

[Bruce Fraser]

I am using Epson 4000 and PK Sharpener Pro with Canon 1Ds files. I print on Epson Enhanced Matte at 1440 dpi. I have several observations and a question.

(1) How many PPI to send to the printer? I was concerned about this, but not any more. I was told by an experineced professional who should know (not at Epson - so far they have failed to answer such questions) that the printer throws away anything more than 360 PPI, so if the file has more than this at the intended print size, better to down-res with bicubic sharper in Photoshop because it does a better job than the Epson printer driver. So I tried BOTH: (a) leaving the file at 451 PPI with no downrezzing, PK Output Sharpener at 480; (b) dowrezzing to 360 in Photoshop and PK Output Sharpener at 360. The prints were 6 * 9 inches image size. Looking at these results with no loupe, there was no detectable difference. Using a 10X loupe there was a small positive quality difference in the mosre minute details in the 360 print. So perhaps this means that for finer detail on big enlargements the advice I got makes some sense.

It is untrue that the printer driver throws away anything more than 360ppi. The printer driver doesn't throw away anything! The base resolution of the print head is 360 DOTS per inch, and that's the basic sieve through which it passes the pixels. However, we have seen a small but not nonexistent benefit to sending 480 ppi, particuarly for smaller-size (less than or equal to 8x10) prints, which is why we went to the bother of building a 480-ppi sharpener. The difference you're seeing is probably equally due to applying a 480-ppi sharpener to a 451-ppi file as to the printer behavior itself.

(2) I have achieved commercial quality results from the Epson 4000 sending ANY PPI SPECIFICATION in the range of 180 to 480 to this printer from A3 size downward. Much depends on the image content also.

I would agree with all of that.

(3) PK Sharpener Pro Output Sharpener seems to break down completely if the print size is very small (e.g. 4*6 inches) and the resolution very high (e.g. above 550-600PPI). Using the highest setting 480, the result on the monitor is tremendous speckling and very brittle imagery. I know the monitor is not a reliable indicator of the impact of sharpening, but this was so extreme it would obviously be a waste of paper printing it. I therefore assume that for such cases one MUST downrez to use Output Sharpener successfully. Is this a fair conclusion?

Output sharpeners will break when the file isn't at the resolution for which the sharpener was designed, and the further from that resolution the file is, the more they'll break.

However, please make one print before drawing any conclusions. My experience has been that sending more than 500 or so ppi through the Epson driver obscures detail that lower resolutions reveal. I have no idea how a 600-ppi file will interact with any of the sharpeners, but I need to emphasize in the strongest possible terms that looking at the sharpened pixels on a monitor will tell you absolutely nothing about the print appearance. At 600 ppi, a three-pixel-wide halo is smaller than you can see without a loupe, and single-pixel speckles are just meaningless?you'll never see them, but they have an influence on the way that pixel and its neighbours get transformed into dots.

That said, if you want to make a 4x6 print and you have more than 480ppi, our recommendation is that you downrez to 480 ppi, then apply the appropriate 480-ppi output sharpener.

[Mark D]

Bruce, thanks alot for the useful insight and the advice at the end.

But there is something you said I do not understand: "The base resolution of the print head is 360 DOTS per inch, and that's the basic sieve through which it passes the pixels." If you could demystify this sentence for me I think many others too would be thankful. Specifically:

(1) Could you please explain what you mean by "the base resolution of the print-head is 360 DOTS per inch"? What is "base resolution" say as opposed to non-base resolution? How does it differ from the 1440 dpi I am printing at? When you say "DOTS" do you mean DOTS as ink and not pixels as picture elements?
(2) What kind of "sieve" is this?
(3) How does it "pass the pixels"?

Thanks in advance for any further precision you can offer on this.

[Bruce Fraser]

I'm not sure how much demystification I can provide, because the topic is inherently mysterious, but...

By dots, I do indeed mean dots of ink. Inkjet printers lay down ink dots, not pixels.

The print head itself has ink nozzles 1/360th of an inch apart?that's the base resolution. The 720, 1440, and 2880 numbers are obtained by the stepper motor moving the printhead by those smaller increments.

Epson's screening or dithering algorithm is a secret black box, the true details of which are known by two guys in Japan. But to oversimplify hugely, it overlays the pixels on a 360-cell-per-inch grid, and for each cell, calculates how to turn the pixels into dots of ink.

John Panozzo of Colorbyte claims that the 480-ppi phenomenon we've observed is caused by 480 ppi being a magic resolution due to a bug in Epson's driver. I don't know enough to argue the point but I suspect he's right?we certainly don't see any benefit to sending more than 360 ppi to ImagePrint.

I suspect that if I were to add anything further, I'd deepen everyone's confusion, my own included.

[Mark D]

Thanks Bruce - what you've said provides some clarification, and I guess the remainder will remain like the recipe for Coca Cola.

[rhkdiver]

Ok, with all of the above discussion there is still an issue that has and still is confusing me, espeically based up this post.

I'm shooting a D2X. The native image size is 4288x2848 pixels, at 300 dpi, with a document size of 14.293 inches by 9.493.

If I want to print on an Epson 7600 an actual image size of 20x30 inches....should change the document size ( in the PS image size dialogue) to 20x30 ( which would give me a 30wide by 19.925 H print) and maintain the 300 dpi resolution OR should I allow the Epson driver to do the interpolation from 14x9 to 30x20.

When I change the Resolution from say 300 to 420 the pixel dimentions change but the document size stays the same. So what is the best way to deal with a file that you want to print larger than its native document size.

[Mark D]

My experience with satisfactorily answering these questions is to EXPERIMENT and see what works best in my context - it only costs a bit of time, paper and ink (and you can economize on that by only printing a portion of the enlargement that reveals best the information you want). Looking at the size change you wish to make and entering it in Photoshop (Image Size dialogue box) with "Resample Image" turned OFF, PPI decreases from the 300 you started with to about 143. Many people would consider this inadequate, but as I suggested above, test it by sending this information to the printer and look at the result from the normal viewing distance one would use for a print of this size. If you find the apparent definition/sharpness inadequate, my next choice would be to resample the image in Photoshop, this time with "Resample Image" ON, select 240 PPI for resolution (should be adequate for that print size) and select BICUBIC SHARPER. Then print and compare. I'd be very interested to see whether Bruce and/or Jeff thinks this is good advice!

[rhkdiver]

That's very good advice. I can't believe I spaced out and forgot to unlick "resample". However, your comment about using BICUBIC SHARPEN if I have to resize upwards using resample, is contrary to everything I've read when using PSCS Image Resize. Bruce, Jeff and the other Photoshop experts all suggest using BICUBIC SOFTEN when up sizing and sharpen when downsizing. Your thoughts?

[QUOTE=Looking at the size change you wish to make and entering it in Photoshop (Image Size dialogue box) with "Resample Image" turned OFF, PPI decreases from the 300 you started with to about 143. Many people would consider this inadequate, but as I suggested above, test it by sending this information to the printer and look at the result from the normal viewing distance one would use for a print of this size. If you find the apparent definition/sharpness inadequate, my next choice would be to resample the image in Photoshop, this time with "Resample Image" ON, select 240 PPI for resolution (should be adequate for that print size) and select BICUBIC SHARPER. .QUOTE]

[Bruce Fraser]

I'm generally not a fan of uprezzing, but when I have to do it, I use Bicubic Smoother.

143 ppi is a bit low?I usually suggest a minimum of 180 ppi for large prints.

I suggest trying the following test.

Print the actual data at 143 ppi, using the inkjet 180 Output Sharpener.

Uprez to 180 ppi, use the Inkjet 180 output sharpener

Uprez to 240 ppi, use the Inkjet 240 output sharpener

Optionally, uprez to 300 ppi, use the Inkjet 300 output sharpener

If you want to factor in Bicubic, Bicubic Smoother, and Bicubic Sharper, by all means do so but that's 9 prints instead of 3.

I suspect that you'll find uprezzing to 240 is the best compromise, very slightly better than uprezzng to 180, and that uprezzing to 300 won't buy you anything you don't get at 240, but I could be wrong.

[Mark D]

You and Bruce are of course right. I got confused about the direction of pixel dimension change between the PPI change and the dimension change. B-smoother is the way to go as this would be up-sampling for sure.

[Mark D]

Bruce or Jeff: grateful if you could clarify the following: the Inkjet Output Sharpeners are graduated 180-240-300-360, then it jumps to 480 with nothing between. For a number of my images, depending on the cropping and image size I end-up with, unless I resample, PPI can be anywhere between 360 and 480. For example in the range of 420, one is "equi-distant" between 360 and 480. Is there any reason why you didn't build a step between 360 and 480, and what do you recommend in these situations?

[Bruce Fraser]

Bruce or Jeff: grateful if you could clarify the following: the Inkjet Output Sharpeners are graduated 180-240-300-360, then it jumps to 480 with nothing between. For a number of my images, depending on the cropping and image size I end-up with, unless I resample, PPI can be anywhere between 360 and 480. For example in the range of 420, one is "equi-distant" between 360 and 480. Is there any reason why you didn't build a step between 360 and 480, and what do you recommend in these situations?

480 is a "magic number" for the Epson inkjets, and is only worth using when you have 480ppi of real (i.e., no uprezzing) pixel data. In your situation, I'd downsample to 360 via Bicubic Sharper, then apply Inkjet 360?there seems to be no benefit to intermediate resolutions between 360 and 480.

[Mark D]

Thanks for clarification Bruce - much appreciated.