« Random Excellence: Henri Huet | Main | 'Tuesday Tips' by Alex Garcia »

Thursday, 13 May 2010

Comments

When I read your last post about this issue, I was dubious. I tried it anyway and I was flabbergasted to see the visible results on my prints. I agree that the excessive pixels result in diminishing returns.

Pixel Genius’s sharpening program has settings for (among others) 300 – 360 – 480 so I settled on 480 for my work. Thanks for turning on the light.

Ctein:

Excellent and useful as always. I'm using an old but sweet Epson 2200, and usually print at 240 to 300 dpi - I've found that going much higher results mainly in higher ink usage and slower print times.

What I have found, though, is that - for me at least - turning OFF the 'high speed printing' option in the printer dialog makes a big difference.

Steve G
Mendocino, currently in Florence

"These two illustrations, above and below, show highly magnified sections"

Just out of curiosity, how high was the magnification? Or to put it another way, how large a print would one have to examine to see the differences in your examples? It seems unlikely that these differences would be visible in a full-frame 4x6-inch print. They would be below the threshold of unaided human visual acuity. So at what degree of magnification would they become visible? I wouldn't want to fret over details that, for all practical purposes, no one could see anyway.

Thanks for doing this Ctein. An informative experiment. I actually performed a similar experiment when I first purchased my Epson 3800 a few years ago...and I promptly forgot about its results. Your experiments reminded me of the experience, but not the results. Now I have to do it for myself all over again...or just take 450 ppi and call it a day!

I've seen very large prints (~3m long) of some of my own images printed by a commercial printer at 72ppi. I was extremely sceptical about how they would look, but in fact viewed from a reasonable distance they were excellent. Since seeing this, I've several times experimented myself with print resolutions of 240, 180 and even 120ppi when needing to make large prints from 12mpixel originals. Of course if you put a lupe on these you may be disappointed; from a "normal" viewing distance they hold up extremely well. 360ppi is pretty good, though - have to admit that!

Have tests been made on black and white laser printers in order to see which models are the best for photos?

the week (or two) after, an interesting take on the question, "How big can I print this?"

I look forward to that one, Ctein. Am I to assume you'll be addressing this issue when using inkjet printers? When printing in labs I imagine it's a different set of rules that apply.

These images say one thing to me: thank god for quadtone black and white printing.

I think any discussion of PPI and DPI in this way "Resampling a 450 ppi photograph down to 360 ppi will cost you sharpness and fine detail." will only continue to mislead too many photographers and designers into thinking PPI or DPI as some type of absolute measure instead of an arbitrary assignment.

From my experience in dealing with designers and even other professional photographers it's amazing to me how many assume PPI is an absolute measure and don't understand it's simply a tag inside the file. The pixel dimensions are the only true absolute measure of course.

So, if I understand what your saying here; if you want great sharpness, say at 900 ppi, then your 12 megapixel camera is great for 3 by 5 prints.

Regards,

Robert

Interesting. Not sure I get all this, but much food for thought.

Rather be in a dark room with wet hands, though.

I see the difference in resolution as you step up; that's quite clear. So here's a question: if you have a print size you require, and you can put, say, a native 300 dpi on it, what do you lose/gain by doing a modest uprez to 450 dpi? I would assume you'd get more apparent sharpness, but would lose..what?

Dear Steve,

File resolution has no effect on ink consumption whatsoever. Only printer resolution settings do. Epson told me that at the highest quality setting for the 2200, you would use about 15% more ink. I had a 2200 and printing at the highest quality setting with "high speed" turned off made a huge difference in the quality of the prints. The impact of "low speed printing" is much smaller with the next generation 2400 and it's essentially invisible with the 2880.

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

Dear Gordon,

As I said in this and the previous article, these were 8 x 10 prints, and I'm reporting perceivable differences with my naked eye. Your mileage may differ.

The sole purpose of the illustrations here is to show that there is a steady improvement in quality of rendering as the file resolution increases. The magnification in the enlarged illustrations is irrelevant. It's simply high enough that the reader can see the individual ink droplets, assuring that they can see every bit of information that the printer was conveying.

Up to 8.5 x 11 or even 11 x 14, most people will peer at a print at their normal comfortable closest focusing distance, if there's fine subject detail that engages them. Up to that point, your results are likely to be independent of print size. Above that, people tend to move back from the print.

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

Dear Herman,

Not by me. The point of these two articles is not to tell you how your printers will perform, it's to tell you how YOU can test them to find out how they'll perform for YOU.

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

Dear Miserere,

Yes, inkjet. I can print a lot larger on an inkjet printer that I can in the darkroom. 'Sides, I'm not terribly interested in darkroom printing these days.

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

Dear Peter,

That is not message you can take away from these illustrations. They say nothing at all about how quadtone B&W and cCmMYkkK B&W prints compare in appearance. Do not try to read conclusions from article illustrations that they are not intended to yield


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

Ctein

This corroborates my non-scientific testing. A 1Ds3 file at 600dpi native resolution through my Z3100 looks fantastic (a nice 9 by 6). Bigger prints still look very nice, but there is a clearly perceivable difference. There is also a very clear difference between a 1Ds3 print and 5D print at 9 by 6 or 10.5 by 7, even though the naysayers suggest you can't tell any difference up to something bigger than 18 by 12.

As a point of reference my eyes are far from 'golden' and I'm not particularly aware of being really fussy.

Mike

I believe the magic number for your Epson R800 is a divisor of its native optimized max dpi, 1440, i.e. 288ppi. I was told this by JP Caponigra, the late Bruce Fraser, and Graham Nash, as well as a few other people who met the "white gloves" of Epson Japan who actually make these printers.

300, 360, 450, 600, 720, 900, and 1200 for the images you posted are nice and round, but they are not integers of anything related to the R800's native resolution, and thus worse than 288ppi, because they are interpolated by the printer.

That's my understanding anyway.

Dear Yunfat,

Errrr, "1440" is a very agreeable number; it happily divides evenly by all integers from 2 through 12, save for 7 and 11. 360 = 1440/4; 288 = 1440/5. So there's a whole bunch of "magic resolutions" that are integral fractions of the native dpi.

Now, that said, there *might* be something special about the 1/5th fraction that isn't true of any other, so I generated another test image at 288 ppi, printed it out, and compared it to the 300 ppi print. Guess what? They are essentially indistinguishable -- hardly surprising with only a 4% difference in resolution. In no way does the 288 ppi print look superior.

In all fairness to Bruce, et al., much of this "wisdom" dates from the era of the Epson 1280, nearly a decade ago! What these folks said might very well have been true at the time, and might still be true for very old printers. What I can tell you is that it hasn't been true for the Epson 2200, 2400, 2880, 9800, or 800 printers, with either Windows or Mac OS drivers. Except in certain pathological cases that have no bearing on real-world printing, magic resolutions don't improve print quality for these machines.

Now, would this also hold for the HP and Canon machines? I have no idea! This is why I'm telling people how to run their own tests. A handful of sheets of paper and an hour of your time will answer any of these questions for your particular machine.


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

The contrast between the experimental approach and the appeals to authority is interesting.

Dear Jeff,

There's really nothing to "get." All I've done is provide you with a simple and robust way of determining how much image file resolution matters to you and to your printer. It's nothing more than that. This column was merely a follow-up to the first one addressing a reader's question about whether printing at any "magic" resolution would produce different results. Bottom line, for the printers I have? It doesn't.

So if it's really all that simple, why have I spent a whole column on it? Because if I merely asserted it, there would be a whole bunch of people arguing that I was wrong or that so-and-so had said on such and such website that I was wrong.

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

Dear John,

I'm not actually sure that you would get more apparent sharpness in the print by upsampling the photograph to 450 ppi. It would depend on how the rendering engine and the printer driver actually massage the data. My instincts have been that unless I'm printing so large that I think pixelation might be visible in the print (that is, I'm at 200 ppi or lower), I don't bother upsampling.

If one wants to move into the rocket science regime, there are some interesting games one can play with upsampling and deconvolution routines that produce genuine gains in the quality of the fine detail. But it really gets down to splitting hairs ( literally, if you're talking about a picture of hair). 99% of the time I do not bother with such games, and they are way beyond the scope of ordinary high-quality printing.

Possibly related to your question: I did three columns on various upsampling methods and programs like Genuine Fractal and Blowup several years back. Here are the URLs:

"It's Bigger, But Is It Better? Parts I - III"

http://tinyurl.com/28wkub

http://tinyurl.com/26mybt

http://tinyurl.com/66rgwp


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

Dear Robert,

Oh, I can't tell you how many times I've been asked to provide a JPEG for reproduction for some publication or another and what they've told me that what they want is a "300 dpi" file. As if that tells me how big an image they actually need! I just about always have to get back to them and say, "That's real nice, but are you going to be running it postage stamp size or full cover bleed. Because the file size I need to provide you is 300 times the dimensions in inches of the size you're going to run that photo at." Then they get the idea, and they tell me what I need to know. I'll bet you a bagel, though, that the next time they call someone up asking for a JPEG for reproduction , all they tell them is "300 dpi."

That said, I think you may be confusing resizing with resampling. When you resample an image, you change its absolute pixel dimensions. In the context in which I'm using it, it is a measure of physical file size, and my usage is correct and non-misleading. I'm talking here about holding the output size constant while changing the ppi, not changing the ppi and letting the output size vary in step with that. And this must be talked about in terms of ppi, not absolute pixel dimensions, because ppi is the measure of resolution/sharpness, and that's what these two columns are all about.


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

Will,
Ctein is simply advising people to run their own tests and see for themselves. But, for better or for worse, 50-odd years of reviewing experience between us strongly suggests that most people won't do so--they'll simply take the author's word for his conclusions. Hence the contrast of which you speak.

Mike

P.S. I've been reviewing for 22 years now. Ctein, you?

As someone else who's also had the "300 dpi" request more times than he cares to count... Isn't the temptation to send 'em a postage-stamp-sized image (at 300 dpi, of course) almost overwhelming? :-)

"Almost", because the desire to get paid is greater.

Some day I'm gonna do it, though...

Mike,
That's what I thought. And I too, am likely to take his word for his conclusions, at least for now. One of the things I most like about Ctien's articles is that he always explains his methods. Even if I can't run the experiment, it's nice to be able to check his work and see if his sums come out.

Will

Personally I'll take Eric Chan's advice about printing with my 3800.

Take a look at what he says about file resolution relative to driver settings which you don't mention much here.

http://people.csail.mit.edu/ericchan/dp/Epson3800/faq.html#native_res

Most of my prints are made from scanned 6x7 an 4x5 film. I learned my lesson the Ctein way, actually seeing a difference in a pair of B&W prints (Epson R2400). I had been lazily using 300 PPI for everything, and then one day I ran across a print that had what looked like a fingerprint in an area of a print. I examined the neg - no fingerprint. I rescanned the neg - same results. Then I examined the image and saw that there was a small fine-grained pattern in that part of the image (acoustical ceiling tile). I changed the resolution of the file to 360 PPI and lo and behold, the "fingerprint" disappeared. What I think was happening was some sort of moire pattern created by a fine clash between the tiny pattern and the non-native resolution I was using. Ever since then, I've gladly used 360 PPI even if there are no fine patterns in the image. So for me, 360 for everything.

mark roberts: send them ten images of various dimensions, but all at 300ppi. Or twenty. When asked why, explain innocently that you didn't know what size they wanted.

I recall questioning a picture editor over this once, and being informed that the image size didn't matter because their software could make pictures bigger or smaller while retaining "300dpi quality" so long as that was originally supplied.

I did get her to say what resolution they liked negatives to have been scanned at, but again, she didn't think the physical size of this negative had any bearing "so long as the 2400dpi quality was there".

Eventually I got her to admit, they were in practice going to just use the picture bigger or smaller depending what they got! What a waste of time even discussing it.

Just a quick note on RIP... the resolution does matter!... RIP's such as that by Ergosoft does have a very large difference in quality at higher resolutions due to the nature of their screening technology... this holds true for my 9900 and my d'Vinci printer... but with the larger resolution so comes bigger files that in many cases are too large for most users and the higher resolution should be as much as possible optical data...

Ctein,

Thank you, thank you, thank you . . .

. . . for such clear and thorough explanations.

McD

Thanks for running the test. I was anxious to hear your results. This kind of esoterica needs to be explored more often, as there is very little information from the manufacturers themselves w/regard to achieving the best output.

Along the same lines as the people requesting "300 dpi" images, I have seen numerous occasions of people saying that web images are no good to use because they are 72dpi (or that setting them to 72dpi is a way to prevent miss-use). That prompted this discussion :

http://www.naturephotographers.net/imagecritique/bbs.cgi?a=vm&mr=42964&CGISESSID=8cd68d573cae39f2ad56f37c34edbb79&u=26105

which is pretty good, because I actually learned a few things - like the origin of the "72" itself!

Dear Jim,

Congratulations (well, that isn't exactly the right word); you're the first person I've run into to encounter one of those pathological cases in real life.

I certainly can't see any problem to come from INCREASING your scan resolution to hit one of the magic resolutions. Mostly, though, I would wonder why you're scanning your film at such low resolutions at all. I've been saddled with having to do film scans for 15 years, ever since I built my Online Gallery. Each time I did film scans (or had them done by a professional lab) I worked at the absolutely highest quality I could ever imagine needing, so I wouldn't have to scan the film again in the future. As you've observed, it is very time consuming and there are better ways one can apply one's resources.

How well has that worked for me? Not so much; I'm currently on the third round of scans of the same photographs! In other words, twice I've been wrong (so far!) about how high a quality I would need in the future. So, today, I scan my medium format film at the maximum resolution my scanner will support, which is 4800 dpi. Do I currently need 4800 dpi scans for very much? No! I've just had the life experience of knowing that I'm lousy at predicting future need. Storage media are cheap. Make big scans for archiving, sample them down to a workable size for your actual printing, and do as much work as you possibly can in adjustment layers. That's my prescription.


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

Dear Will and Mike,

OK, you sent me scurrying to my brag shelf -- first published article was in July of 1977 and first product review was in July of 1981, so Mike's estimate of 50 years reviewing experience between us is on the mark.

As for appealing to authority versus experimentation, 99% of what I know comes from authority. When I read Science or Nature, do I actually go and run the experiments they report on myself? No! I don't have the time or money, and in fields outside my discipline I don't have the skill! If I read any survey or summary articles there, which doesn't even report experimental results, do I disbelieve merely because someone is writing from their authority? Absolutely not. Not without damn good reason. And damn good reason has to be more than "He disagrees with my preconceptions."

Which is probably what separates me from most of the human race, which seems much more inclined to reject something they've been told, regardless of the authority of the source, solely because they believed something different beforehand.

I would also add that most people on the planet aren't capable of running proper experiments. There's a specific skill and knowledge set involved, as well as training. It's simply not possible for most people to verify what they are told, even if they had infinite amounts of time and money to do so.

The reason I'm especially tickled with the articles I've done here on print sharpness is that I came up with such a simple and robust experiment that pretty much anyone could do it. That's hard! I think I am extraordinarily clever, if I do say so myself [ grin ].

That said, my track record, established over hundreds and hundreds of articles, is that I'm right at least 99% of the time. Which means, if someone comes to me and says, "prove what you're telling me is correct," my reaction is, "no, you prove what I'm telling you is wrong," because the odds are strongly on my side.


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

Dear Don,

Meaning you're going to run some tests prints and experiment yourself, right? Because that's what Eric recommends that you do. He points out that what he's describing is logical but he hasn't run the tests to confirm that it works best.

It's a nice webpage, by the way. I'd recommend it to anybody.

On a related matter, he makes explicit an assumption that I've always wondered about. People assume that it's better to do the "interpolation" in Photoshop (or whatever) than to let the printer driver do interpolation. Other than the rare pathological cases that have been discussed, does anyone have any EXPERIMENTAL data to support this? Or are people just assuming that Photoshop will do a visibly better job? The answer is not intuitively obvious to me.

My own experiment showed one clear case in which this assumption fails; although the differences are nitpickingly small, my 1200 ppi print is definitely sharper than my 720 ppi print.

I have generally observed this, although not running the kind of controlled experiment I did in this column. I think there are a lot of people who are assuming that any information provided to the printer above its native resolution is simply clipped. That is possibly true for some printers and driver, but it is not true for any that I use. Rather, as demonstrated by two columns on the subject, it's more like combining sources of blur on the imaging side. Up to a point, reducing any source of blur improves sharpness, even if the blur you're reducing isn't the dominant one.

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

The comments to this entry are closed.