Study of Printer Resolution

It’s nice to come across, now and again, someone on the web who actually knows what the terms dpi, resolution, etc. mean. Digital imaging is widely misunderstood on the net, to the extent that at least 90% of sites asserting the superiority of one product over another come up with findings wildly out of synch with truth and reality.

Nevertheless, I see at least one flaw in your reasoning. You state the following:

"When scanning such a negative it takes a minimum of 2 "dots" or samples to resolve a line-pair. Under certain circumstances it can actually take more, depending on the test pattern, but in practice this is a good rule-of-thumb."

No doubt it is a good rule-of-thumb, but if you’re aiming higher, it’s not very useful. If you scan a vertical or horizontal "line" (which obviously has width on the negative, if not in mathematical theory) of one pixel width (of the scanning device), which happens to have a centre in the division between two CCDs, the two CCDs output equal values (i.e. no contrast) so the line is not resolved. (Forget diagonal lines, which require higher resolutions again). By increasing the resolution of the scanning device, you can reduce the minimum width of the line which can be resolved, but in theory you will NEVER be able to capture all the information on the negative (in practice you can get as close as makes no difference with a resolution of 10000 dpi or so with typical slide films like Provia F). Note that your "certain circumstances" in the phrase I quoted here is of course the rule rather than the exception, but 4 "dots" would resolve the line-pair with acceptable dependability. Nonetheless, you do recognize and state that it is only a rule-of-thumb, so I do not have a problem with this (especially considering it eliminates one more variable for the dummies who can’t be bothered to learn for themselves).

However, despite your saying so, you come to the conclusion that, because sending more than 350dpi to the printer isn’t going to increase actual spatial resolution, there is no point in scanning at higher than the scaled up resolution (with anticipated print size taken into account). This is wrong.

According to your findings, when printing an approx. 11 inch image using the full length of the 35mm negative (36mm), you will not see benefits from a scanning resolution of above 2700 dpi. Well, get a negative with plenty of detail (areas with over 100 lp/mm for example), and scan it at 2700 dpi and have the same negative commercially scanned at over 5000 dpi. Print both resulting files to the same printer (which is capable of a spatial resolution of 350 dpi). View the results, ignoring the obvious differences in shadow detail, etc. and concentrate on just the resolution on the prints. The print obtained from the scan at double resolution will contain more detail (not double the detail of course). I’ve tried something similar myself to satiate my own curiosity.

The long and short? Scan at the highest resolution you can possibly afford (higher resolution also increases perceived contrast), and print at the highest resolution your printer can cope with. Don’t kid yourself that your cheap scanner is getting you optimum results because it can scan at a resolution "high enough" for the printer, based on the conclusions on Peter Nelson’s web page.

A few comments:

I'd suggest using the terminology pixels per inch (PPI) when referring to the bitmap resolution. This will help to minimize confusion when discussing printer resolution in dots per inch.

As an owner of an Epson 1270, the assumption about desirable print size isn't as applicable. This assumption also ignores the possible need for cropping.

I have found that overscanning and downsampling produces a superior image to simply scanning at the desired target resolution. When I can cost justify a desktop scanner approaching 4000 pixels per inch resolution, I'll be very pleased.

Thanks for offering up a thoughtful analysis. Well done and well presented!

BJ

Nevertheless, I see at least one flaw in your reasoning. You state the following:

"When scanning such a negative it takes a minimum of 2 "dots" or samples to resolve a line-pair. Under certain circumstances it can actually take more, depending on the test pattern, but in practice this is a good rule-of-thumb." No doubt it is a good rule-of-thumb, but if you’re aiming higher, it’s not very useful. If you scan a vertical or horizontal "line" (which obviously has width on the negative, if not in mathematical theory) of one pixel width (of the scanning device), which happens to have a centre in the division between two CCDs, the two CCDs output equal values (i.e. no contrast) so the line is not resolved.

That's why I said "under some circumstances it can take more". As a practical matter there are very few situations where there will be a noticable advantage to scanning at more than 2X the lp/mm resolution of the system (film+lens) that produced the photo. A good web site with plenty of examples illustrating this is:
http://www.users.qwest.net/~rnclark/scandetail.htm
The author was examining how much detail could be captured in scans of 35mm film. He found that there was a significant difference between 2700 and 4000 DPI scans, but only the tiniest further improvement going from 4000 to 6000 DPI.

However, despite your saying so, you come to the conclusion that, because sending more than 350dpi to the printer isn’t going to increase actual spatial resolution, there is no point in scanning at higher than the scaled up resolution (with anticipated print size taken into account). This is wrong.

According to your findings, when printing an approx. 11 inch image using the full length of the 35mm negative (36mm), you will not see benefits from a scanning resolution of above 2700 dpi. Well, get a negative with plenty of detail (areas with over 100 lp/mm for example), and scan it at 2700 dpi and have the same negative commercially scanned at over 5000 dpi. Print both resulting files to the same printer (which is capable of a spatial resolution of 350 dpi). View the results, ignoring the obvious differences in shadow detail, etc. and concentrate on just the resolution on the prints. The print obtained from the scan at double resolution will contain more detail (not double the detail of course).

There are two possible explanations for this. One is that, as my tests showed, there is no sharp ceiling for printer resolution. Not only is there more resolution on one axis than the other, but there is more resolution for certain colors than others, as you can see in the sample images. So printer resolution doesn't just hit a wall - it "falls off" sort of the way frequency response rolls off on a stereo speaker. In the vertical axis the HP was still able to resolve single pixel lines at 700 DPI for some colors! I should probably add a comment about this in my report.

The other factor about why a 5000 DPI professional scan may look sharper is that the professional scanner may have better contrast, better signal:noise ratio, better registration during the scan or other factors not related to pure DPI resolution.

Roger Clark is obviously quite knowledgeable about this subject, and his is a very good depiction of the digital/film/scanning scene. But even so, his site leaves a little to be desired.

When using the test print (from 4x5) instead of the actual scene for testing, he says:

"If there is any bias in going from actual scene to test print, the image detail in the test print is slightly higher, partly due to the increased contrast."

Is he trying to say the contrast of a Cibachrome is higher than that of a brightly lit scene??? You can only get something like 7 stops of dynamic range on paper, while you can get far, far higher than that in the great outdoors. I’m sure he would agree with me that this is absurd, but leaving statements like that in the final write-up always tends to diminish the whole site’s credence. I wish these guys would proof-read their stuff before uploading to the net.

When comparing the digital to film, he used a cheap digital camera which I assume to have an inferior lens to the 35mm system he used. Also, testing at close range (the print) as opposed to infinity (the real scene) is not altogether fair, as the different systems will cope with the close focus with different capabilities. Also, in the final comparison between digital/film, he continues to assume that film is the limiting factor in resolution, when in fact at those resolutions and Mega Pixel levels, *lens* resolution does become the limiting factor (with films of far higher lp/mm capacities than Velvia, or imaginary films of unlimited resolution).

But, to the purpose of your referral to this site: you say he found only the "tiniest" improvement going from 4000 dpi to 6000 dpi, but I don’t think that’s exactly the way he put it:

"There is a significant increase in detail from 2700 DPI to 6000 DPI 35mm, and proportionally in between." [my emphasise]

"from 4000 to 6000 DPI... ...grain is resolved better, but that increase is important for big enlargements (more than about 11x14)."

"To some, the above images do not show significant detail above about 4000 dpi in the 35mm scans (I think they do)."

"3 parallel grass blades are resolved... ...at 6000 dpi (maybe slightly less distinct). At 4000 dpi the blades are not fully resolved, just a hint that one might think is grain."

"From the above scans, it shows that 35mm shows increasing detail at least to 6000 dpi in the drum scans"

"(and note that the 4000 dpi scan does not get all the detail on the film)"

"Film people (I've even done it) often cite that they scan 35mm film at 4000, 6000, even 8000 dpi getting additional detail out of the film. This is true."

Nonetheless, I agree with you that scanning Provia F at 8000 dpi for a 8x12 Epson print is a bit of an overkill. :-)

BTW, it was interesting that you found the HP to be clearly better at reproducing fine detail than the Epson.

I've updated the conclusions to give more weight to the possible benefits of higher resolution scanners. I've also added the reference to R.N. Clark's page in the scanning resolution discussion, and I've changed the test pattern resolution description from DPI to PPI to minimize confusion people might have with the printer-manufacturer's own DPI ratings for their printers.

Feel free to offer any other suggestions or comments! Thanks!

Excellent!

as an engineer, I should be up to investigating exactly how the Epson munches data, but frankly .. why bother. seems to me, the driver, which is an unknown quantity, does the dirty work. think about it. you could sample at a billion billion dpi (apologies Carl Sagan) and dump it down the printer. what happens to the billions of unused bytes that cannot map to the printer resolution? gets resampled, data gets tossed, and we end up with exactly what we need to meet the resolution and size requirements. again, without giving much thought to this, if there is a visible improvement from extreme oversampling, I would think it is an artifact of how intelligently the driver downsamples to create the printer image, since the data block size is constant.

honestly, I saw an exhibit today of 8x10 contact prints from the early 1900's, and am of the mind to toss all four of my inkjets in the trash.

I did a little more thinking on this test:

I believe that if you made a diagonal test pattern, the results would be considerably different. There is no way to make a continuous 1 pixel wide line on a diagonal. In fact, the only way to simulate a non vertical or horizontal line is the rely on resolution which is high enough to resolve an image of the line and to allow for anti aliasing techniques. The true test of resolution is in the rendition of curves and non-perpendicular lines. A better test of resolution would require diagonals and perhaps concentric circles in the source image.

BJ

This is an interesting thread, and one with some well intended homework and intensive examinations of both HP an Epson devices.

The issue I have though is how any of this relates to the photographic capability of either of these printers. For continuous tone applications like photographic out-put the printer that has the highest absolute DPI, or fringe resolution doesn't always make the best at making photographs. For that matter, I'm curious how a high-end, color laser printer would stack up in these tests, and we all know how those do for photographic applications.

Since the Epson tends to "blend" straight pixel lines more than the HP I would conclude to some irony that the Epson is the better photographic printer. Indeed, while HP's are great general purpose printers, they do have problems in color transitions which makes them look "grainy" in some tonal transitions, and hence earns their reputation as second best to Epson.

Scott says - "For that matter, I'm curious how a high-end, color laser printer would stack up in these tests, and we all know how those do for photographic applications."

Sorry, not all of us know how colors laser printers perform for photo applications. Please tell us, briefly. Thanks.

{{Sorry, not all of us know how colors laser printers perform for photo applications.}}

The best way to find out for your self is to go to any 24-hour copy center, spend $.50, and have an 11x14 made off a Canon CLC or HP 8500 from your favorite digital file.

I'm just curious as to what the implications and reasons of the above tests are, and if we are concluding from thus that the Epson 870/1270 is inferior to the HP 900 series for photographic use.

The issue I have though is how any of this relates to the photographic capability of either of these printers.

I think the Epson is a better photographic printer, but that wasn't the question being investigated. The question being investigated was what implications the spatial resolution of the printer had for the user's choice of scanning resolution. There are pleny of other factors that determine how well a printer reproduces photographs, and there are plenty of factors other than resolution that determine the quality of the result a scanner produces, but none of those were what my study was about.

FWIW, even though my tests indicate the HP may be slightly sharper in the spatial resolution of its printing, subjectively, I find the Epson to produce far more pleasing results when printing photographs. This is due almost entirely to its much smoother gradation in tones and colors, and almost total lack of any dithering artifacts, unlike the HP which produces a dither pattern easily visible to the naked-eye in light continuous tone regions such as sky and light flesh tones.

I was very surprised that the HP seemed to slightly beat the Epson on spatial resolution because on paper the Epson has higher resolution and smaller dot size. One possibility may be that the Epson, even with its smaller dot-size, cannot place those dots with as much precision. The other possibility is that the Epson was laying down too much ink. The HP driver lets you set the ink volume directly, but the Epson driver only gives you indirect control, by setting the CMY values. But even with the CMY values backed-off to -25 I was unhappy with the amount of ink it was laying down.

Sorry, not all of us know how colors laser printers perform for photo applications. Please tell us, briefly. Thanks.

They're gawd-awful! On the other hand, I'm not sure I agree with Scott that they would do well in the resolution tests, either. The HP 970cse has a dot-resolution of 1200 DPI, and the Epson has a dot resolution of 1440 DPI, although as I noted, this does not correspond to a pixel resolution anywhere near that high. But we just installed an expensive, floor-standing HP color laser printer at my work and I think it's only rated at 600 DPI. Maybe this week I'll sneak some glossy paper into it and try printing my test images on it.

The other classic problem color laserjets have is poor registration. I don't know why this is but I think it would be a miracle if any of the color laserjets at my work could maintain anything like 1/600th of an inch registration for more than a day.

Hey now, Tony who works the graveyard shift at Kinko's Copy Center swears the Canon CLC 800 series in "darn near photographic quality", so who are we to dispute that endorsement :^)

Oh yeah, could HP *please* fix the yellow ink in their printers while were at it? Your trusty box of Crayola's cleary indicate that what the HP 900 series thinks is pure yellow ink is actually bright orange. And, uh, having t print everything in landscape mode and having Photosop doing all the rescaling rather than the printer driver to avoid the dreaded HP "black lines" is a bit annoying.

Peter, I want to thank you for excellent work. I'll add a link to you to my new web page (just finished Saturday and still being revised). My page explains image sharpness, resolution, and MTF in a unique visual way, using a target of increasing spatial frequency and observing the effects of film, lenses, scanners, and sharpening algorithms on the image. I use accurate computer models based on published MTF data. The page is on,
http://www.normankoren.com/Tutorials/MTF.html
I've been scanning happily with a HP Photosmart S20 2400 dpi scanner. It makes extremely sharp 8 1/2x11 prints on my Epson 1270. 13x19 prints are also very sharp and pleasing, but they certainly aren't as sharp as the 1270 can print; you can tell they're not large format. My modeling shows a clear advantage for 4000 pixel per inch scanners, particularly when appropriate sharpening is applied. The importance of sharpening can't be underestimated. You'll notice I didn't include printers in my modeling-- mostly because I don't have data on MTF. You've inspired me. I may follow in your footsteps and make prints with a varying frequency target similar to the one on my page, scan it with an Epson 1640, and see if I can write a program to derive the printer MTF. More sharpness geek madness, but fun. When the target is available, I'll share it. Thanks again.

I've done it! Inspired by Peter's excellent work, I created a printer resolution page,

http://www.normankoren.com/Tutorials/MTF3.html

It contains a 10mm virtual target whose spatial frequency varies continuously from 2 to 200 lp/mm and a Matlab program for analyzing the scan of the target printout. It shows incredible detail about printer performance-- ultimate limits and how sharp prints will be at various magnifications. I'll put up a post announcing it in a week or so.