« Samsung NX 10 Announced | Main | Resolution 2010 »

Tuesday, 05 January 2010

Comments

Hi Ctein,

As a newbie to print making (sort of) this is very valuable information. Not that I would expect anything less from you!

Cheers,

Chris

You realize of course that the driver for the R800 will resample to 720ppi before screening? (360ppi or 720ppi on the 9800 depending on Finest Detail.) Aligning input resolution to the above is more important than the resolution itself.

I wrote an article sometime ago on what I call Print Surface Resolution (PSR)... huge variable is also the screening technology used by your specific printer or printing software (RIP)...

http://figitalrevolution.com/2007/12/07/print-surface-resolutiona-proposal-by-stephen-schaub-2/

Cheers-
Stephen

I'm running a double-blind sorting test of print sharpness. I printed out an image (high resolution stitched panoramic landscape so the initial source image did not limit the available detail) at 120, 180, 240, 288, 300, 360, 576 and 720 pixels per inch, labelled them with a random number on the back, shuffled them and offered them to a mix of people (young, old, male, female, photographers and non-photographers) to sort from "best" to "worst", independently at their leisure in roughly the same lighting conditions. It would take typically around 10-15 mins for each person to make their choices.

Though my current sample size is small (n=14), there are still some trends are beginning to look statistically significant (need to get a few more victims/subjects before I would make any firm conclusions). Practically everyone placed the 120ppi print last. There is a statistically significant difference between the ordering of the 720 and 576ppi prints (with preference going to the 720ppi print). The 180ppi print is significantly worse placed than the 240ppi print. In general, people had trouble with ordering the 240, 28, 300 and 360ppi prints, the current data does not indicate that my test population on average could distinguish between them reliably (though there were individuals who could reliably order all the prints correctly).

Though I'd like more data to make firmer conclusions, it looks as if 240ppi is OK for general consumption, 180ppi is marginal, and 120ppi markedly inferior. However, I could tell the difference between 720 and 576ppi so I know that printing at greater than 240ppi won't be wasted if the image is suitable.

Printer was an Epson R2400, paper Ilford Gallerie Smooth Pearl, 2880dpi.

Disclaimer: This is an informal study and there are many things which could certainly be improved (e.g. range of test images/papers/inks/printers etc., standardised bright viewing conditions) but its results so far have been useful as a guide as to the region of "about good enough", at least for highly detailed landscape imagery.

"Traditional printers should not be feeling smug; you can't do it in the darkroom, either, unless you're making contact prints from very high quality negatives."

Smug commenter here. My prints are all contacts (5x7, 6-1/2 x 8-1/2 and 8x10) made from negatives exposed at apertures no smaller than f/45. Every one meets your 30 lp/mm "perfect sharpness" criterion.

As a bonus, the darkroom is much more satisfying to me than digital imaging too. :-)

My Great-Grandfather Frederick Hutchenreiter picked the hobby of photography near the beginning of the 20th century. Most of these images were snaps of family, possessions, interiors and the American West- Yellowstone, The Grand Canyon Etc., the family lived in Wyoming. I took up the task of scanning these and other family photos about 95 years later. The negatives are no longer available, only the prints survived. I am thankful that the prints were made by professional studios and are of a resolution that exceeds a "good enough for viewing" standard as the prints were small enough to store and distribute, yet of a high enough quality to enlarge these hundred years later. The details of life a century ago as my ancestors lived, captured in this scale are a treasure to me (probably mundane to them). It makes me wonder what our Great Grand Children might do with snaps taken this early in the 21st century.

Dear Stephen Best,

I have heard many people say this, and I consider it to be pretty much useless advice. In real-world photography, one does not get a say over the native resolution of photographs. For example, if my camera produces a 3000 x 4000 pixel photograph and I want to print it out with one-inch borders on 11 x 14 paper, then I'm going to be printing at 300 PPI. The only way I can print at 360 PPI is to either upsample the image (which will do nothing to improve the quality) or make the print about an inch smaller (which is not what I want to do).

In the few experiments I have done with synthetic images that I could artificially and arbitrarily rescale, I have found this business of matching the magic 360/720/whatever ppi to be of very little import. Indeed, there are single-pixel-scale artifacts that appear in extreme cases at non-magic resolutions. They are situations that are extremely unlikely to show up in real-world photograph, only in one of my torturous test targets. Furthermore, in real-world photographs, 300 PPI looks a little less sharp than 360 PPI, which looks a little less sharp than 420 PPI, which looks a little less sharp than 500 PPI, et cetera. Fundamental resolution of fine detail has a far bigger effect on image quality than matching magic numbers.

I certainly haven't done definitive testing on the subject; you may very well have found a situation where this matters. Me, I have done enough to convince myself it is not worth pursuing. Especially since I can see no practical value to the result, outside of scientific imaging situations where it is important to keep artifacts as close to zero as possible.


~ pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

Dear Daniel,

Those results are entirely consistent with what other people have found, but assuming you're continuing running this kind of test, I would suggest you eliminate the three prints between 180 and 360 PPI in favor of one print midway between. It is extremely difficult to see resolution differences of less than 20% except under carefully controlled test conditions. I would not expect most viewers to be able to distinguish between 240, 288, and 300 PPI photographs. You need bigger intervals to find out if the additional fine detail really matters to your viewers. All you're demonstrating in that range at this point is that they can't see slight differences in fine detail (what researchers call a just distinguishable difference or JDD).

You have the right attitude about this. It's not about finding out what will stand up to scientific scrutiny under standardized conditions, it's about finding out what works for you and works for your audience.


~ pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

I'm sure I'm missing something obvious, but I'm not sure I understand the difference between the Figure 2 and Figure 3 images. Is Figure 2 the image as you'd see it on the monitor and Figure 3 the same images as they are rendered by the printer?

Hey Ctein, thanks for such a fun exercise on a gloomy, bitterly cold winter day (here in Chicago)! I plan on taking a swing at it Wednesday. (We're expecting a blizzard in the evening.)

As long as we're assessing our printers and our visual acuity here's a kindred test, this time for your color vision accuracy courtesy of the X-Rite Company.

Hmmm,
This could well be the dot in the dark!

Dear Douglas,

Exactly!

pax / Ctein

"The only way I can print at 360 PPI is to either upsample the image (which will do nothing to improve the quality)"

How good do you think resampling in the driver is? It's not necessary to resample to driver resolution, but to a resolution that is synergistic (and arrived at by testing).

Presumably you're familiar with the following:

http://www.ddisoftware.com/qimage/quality/

I think we should take again the analogy with audiophile world that you already used in previous articles.

Quality (and size) of print depends ultimately on the subject. I saw that some mythical pictures (on original print) were very poorly defined. I think particularly of Cartier-Bresson, Walker Evans... This was mostly due to the choice of 35mm film, and the conditions of shooting. Were the images weak ? Certainly not !

On a modern side, all the Lomo movement takes its strength from blur, color shift etc... It bear a real parenthood with the musical lo-fi movement.

On the other side of the scale, I've recently seen some massive prints (2m x 4m) of industrial landscape in very high def (Chen Jiagang "Third Line"). You were almost drowned into the image. Would they be weaker at a lower def ? definitively !

My generic feeling is that landscape work gains from the highest definition available, while street photography would be much less demanding. Of course, you can find counter examples.

Just like Classical music is much more demanding on HiFi gear than noisy rock or techno (no judgement of value here, I do like techno)

So at the end of the day, would I need better, more defined, with more depth of color ? I would say that it depends, this is an artistic choice.

Like a previous poster I made a number of prints between 180 and 720ppi and asked experienced photographers to rank them. Their preferences were markedly different based on the subject matter of the photographs.

As a general rule detail rich cityscapes needed high ppi's to fully satisfy viewers, portraits needed low to medium ppi's, and detail poor landscapes needed only low ppi's.

I came away from that experience believing that most photographic resolution debates are meaningless unless the subject matter is fully considered. For example, the viewer looks to identify faces or find sufficient textural information to identify materials. But once these practical visual needs have been satisfied the viewer seems to place little if any premium on additional resolution.

I'd be a little concerned about the ordering of your process, Ctein. By doing the pixel peeping FIRST followed by the more subjective 'which print do you like best?' step, you're more likely to have learned what the "right answer" is. I'd just swap the two steps around.

A thousand questions just sparked up when reading this post and the comments - e.g. "What happens when you sent the printer a file with a higher resolution than it can print, and vice-versa?"
Anyway, thank you Ctein. Shows me that I understand nothing about printing. Time to read a book - which I'll buy over Mike's amazon links...

For the last 12 years I've owned a digital fine art printing business and the one question every customer invariably asks me is "how big can I print this?" which is sort of the real world equivalent of the question you are addressing.

My answer is always equivocal. It depends on circumstances (viewing distance, subject matter, lighting) but, more importantly, it is very subjective. Everyone evaluates photographic prints using different criteria and standards of what's acceptable. In my experience, non-photographers by and large are much less concerned with the technical characteristics of a print and much more concerned with the emotional impact of the image. People experience the whole "gestalt" of the image, not individual traits of the print.

The most important thing is to produce a print that expresses as well as possible that emotional impact to it's intended audience. There are lots of choices involved in this - to be honest, printer ppi is one that I spend relatively little time on as I find there are many others of much more import.

I've done this sort of test for each of my pro-level printers before putting them into use, and found that the results were quite similar in my eyes with all topping out at around 720. While I have no data to back it up, I did set my test ppi test resolutions to be as mathematically simple as possible (.5x, 1x, 1.5x, 2x....based on 360 ppi) to prevent any errant pixels caused by mathematical errors during resampling. If nothing else, it made me feel better.

Dear Andrew,

That's the reason for labeling them on the back, not the front. At a first glance, all the different resolution prints are going to look identical, especially the higher-resolution ones. You won't know what the "right" answer is. The reason for pixel-peeping first is to eliminate any ultra high resolution prints that DON'T present a significant difference. For instance, if I had also run a print at 2400 PPI, I'm certain it would resolve no more than the 1200 PPI print. Including that in the analysis wouldn't serve any useful purpose; I would be straining my eyes trying to look for a difference when one didn't exist. In some experiments, that can be a useful control; in this case it's actually more likely to produce spurious results because you're straining to see something that isn't even there. It distorts the way you look at things and introduces a certain amount of perceptual "noise" into the system. You're not trying to trick yourself or your viewers, you're trying to find out if there are meaningful differences that can be seen.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Dear Stefan,

If you try to print a file that has much more resolution than the printer can reproduce, all you do is waste some CPU time. Bigger files take longer to render, but the printer doesn't care what you throw at it. It will print to the best of its ability.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Dear Gary and Guillaume, et al.

Yes, the results are very content dependent! That's why this is a good, robust test; you can load your test folder with sample photographs of all different sorts that matter to you(as I recommended) and you'll be able to see in the contact prints where the extra resolution makes a difference and where it doesn't. In the sheet reproduced in figure 1, there are a few photos there where there isn't even a viewable difference between the 300 and 600 PPI prints. There are lots where there is no viewable difference between the 600 and 1200 dpi prints. Pixel peeping, ala figure 3, would show that the printer has rendered increasing amounts of detail in all those photographs with each jump in resolution, but I just can't see it in "unassisted eyeball" viewing.


~ pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

Dear Stephen,

Ah, now I see where you're coming from, and I think we are probably in accord.

I am forced to say, though, that I am not familiar with Qimage. We are of different religions [g], and so we are fated not to meet. I am certainly aware of it, but I could not say we are on "familiar terms." But you are hardly the first person with a discerning eye to recommend the program to me; I have heard worthwhile things about it from many people whose eyes I trust. I would recommend it unreservedly.

I would not argue at all against the assertion that a quality third-party RIP, used with the resolution settings that optimize it for one's printer, produces superior fine detail, all other things being equal.

That's different from saying that one should upsample or downsample one's photographs to meet a magic resolution number. I would still argue strongly against that. I suspect you would, too.

As a caveat, I would mention that the test target that Digital Domain provides is NOT a good one for doing the kind of evaluation I've discussed in this article. It's precisely the sort that I warned against using, because it's designed to ferret out anomalies in the rendering engines. Those are very useful tools for analyzing what's going on in a system; I've got a few that are much more diabolical than that one that I can trot out when I'm doing product tests. But it will not give the average photographer a proper assessment of how well their printer will work for their photographs.

Another caveat for casual readers: note that the comparison tests done on the Qimage page were done some years ago with several generations-old printers and software (e.g., Photoshop 6 and an Epson 1270 printer). Don't attempt to generalize the specifics to current systems.


~ pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

Thanks a lot! You just blew my justification for buying that new DSLR.

BTW, in "analog" B&W printing, the limiting factor in resolution is usually not the negative, but lateral halation when exposing the print.

Dear Bob and Chuck,

My feelings on this matter are much the same as yours. Crudely put, I looked at it so I could forget about it. Like almost every other aspect of photography, this is not worth obsessing about. Once you have a ballpark idea of what your printer can do and what you care about in a print, which are two different things, it's time to move on.

You would be amazed and appalled, though, about the amount of verbiage online on the subject. Printer-pixel-peepers are even worse than the noise fetishists, sometimes.

That's why I thought it important to write this article: so the saner folks can learn as much or as little as they want about this subject for their very own selves and then move on.


~ pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================


Dear Ken,

I don't know what "lateral halation" means, but the print paper's resolution is almost never the limiting factor on print sharpness.

This very question produced a hot and heavy debate in the letter column of PHOTO Techniques magazine nine years ago. The then-editor (some guy you never heard of) charged me with the task of sorting out fact from fiction. I ran resolution tests on a large number of black and white papers. With one exception, which I'll describe below, all papers resolved 65-140 line pair per millimeter. It is true that a very good black-and-white negative can have a resolution within that range. So, if you're contact printing, you could be paper-limited ... in theory. In practice, it requires an extremely high pressure contact printing frame to ensure intimate enough contact between the negative and the paper emulsion to get the resolution figures mentioned above, far more pressure than a typical spring-loaded frame can provide. Without very high pressures, there remains a small gap between film and paper and the image is blurred more by traversing that gap than by any inherent limitations in the paper. Odds are, you are never hitting the resolution limit of the paper with anything you print, not even close.

The one exception is kind of an interesting puzzle; an Ilford paper expert alerted me to this anomaly. Variable contrast papers produce much lower resolutions at low contrast than high contrast. (I'm not talking about MTF here; I'm talking about limiting resolution.) Ilford Multigrade IV exposed to pure green light resolved only 50 line pair per millimeter. The same paper exposed to pure blue light resolved 125 line pair per millimeter. Both Ilford and I saw this phenomenon, so I truly believe it's real and not an experimental artifact. Neither of us can explain it. The difference in wavelengths between blue and green light is not sufficient to produce that big a difference in resolution. Rayleigh light scattering goes as the fourth power of wavelength, so that would be a large enough difference, but scattering INCREASES as the wavelength gets shorter. In other words, if the drop in resolution were due to Rayleigh scattering differences between the wavelengths, then the blue light image would be blurrier. There is another kind of scattering called Mie scattering that does increase for longer wavelengths, but it doesn't go anywhere near the fourth power of wavelength.


~ pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

Once you've trained your eyes to discern the subtle nuances of prints output at various ppi under magnification, try the same test with your prints matted, framed, and under glass, having hung them on a wall illuminated at 9-12 foot candles by lamps with no less than a 100 CRI.

Dear Ctein (and all),
you wrote relating to the double-blind experiment :
> You're not trying to trick yourself or your viewers, [...]

That bit made me think...
Isn't sharpening, something generally accepted as necessary for printing digitally, basically a plain old trick?

Wouldn't be interesting to throw output sharpening in the equation - sorry I'm thinking out loud but a stronger output sharpening might make moderate-size detail look better at "normal" viewing distance, while actually obscuring finer detail?

Nice. I did some testing on my Epson 3800 as well with respect to the various high speed, 1440/2880 and finest details settings. For the papers I used (arctic polar lustre), high speed at 2880 was a sweet spot for proofs and 2880 at low speed was incrementally better and what I use for customer prints. The finest detail setting makes the most difference when printing logos or text. Surprisingly the "worse" settings would still be more than acceptable for most folks I suspect.

Dear Nicolas,

Sometimes output sharpening makes digital prints look better, sometimes it makes them look worse.

The "general acceptance" that it is necessary is wrong.

Personally, I am not a big fan of 'throwing things into the equation;' it rarely produces a more useful answer. Usually it just muddies the results.

You can spend forever testing and obsessing about just what kind of sharpening, and how much of it, works best for what kinds of photographs. Bored now! Not my cuppa.


pax / Ctein

Dear J,

The purpose of this test is not to "train your eye," it's to find out what works for you under your chosen viewing conditions. As I said,

"Now, understand that "good enough" is going to depend upon subject matter, viewing distance, choice of printer and paper, and printer settings. Everyone's going to get a different answer. The nice thing about this test is that it takes all of that into account. It comes up with answers that will work for you.... ... Just look at the prints. You get to decide what distance you're viewing them from, what the lighting conditions are, and how persnickety you want to be."

If the conditions you described are ones that you're sure your prints will be viewed under, they're right for you. They'd be wrong for me.

pax / Ctein
==========================================
-- Ctein's Online Gallery http://ctein.com
-- Digital Restorations http://photo-repair.com
==========================================

On the subject of how much sharpness is good enough (which is different from how much resolution is good enough), at the Minneapolis Institute of Arts today I had a close look at Vermeer's painting, The Astronomer. I was intrigued to see that close up the definition is quite soft and at least some of the highlights are quite blobby. Maybe edge sharpening is not so important after all!

Ctein-

err, I'm joking, sort of... These are not my recommendations, but those of the conservators I work with. This would be the usual lighting setup in a museum or some galleries.

Actually, paper/photo conservators recommend zero foot candles for illumination (ie. climate controlled storage), since their viewpoint is that "art is meant to be preserved, above all else"- that includes having people look at it, cuz light kills. Not my view, but theirs.

My personal preference is to avoid looking at anything over 72dpi. And it has to be backlit.

Peace out ~J

The comments to this entry are closed.