Printing from 4000dpi vs 5400dpi scans - at which print size do you usually begin to see a noteworthy difference?

A question for the digital printing gurus here.

I've recently started sending some of my scanned B&W negatives to a lab to have them wet printed (see this thread for context).

For my scans from 35mm, I've been using my main film scanner, which operates at 4000dpi.

I also have another dedicated film scanner operating at 5400 dpi (according to tests around the web, it's more like 5000 real dpi) which has been sitting in storage for a while.

I was wondering if someone could refresh the theory for me and also contribute their personal experience on the ballpark print size one would have to aim for to start discerning a difference between a 4000dpi and 5600 dpi scan of a full 24x36mm negative.

Or - looking at it from the other side - is it worth unpacking and setting up the 5400dpi scanner if all I'm aiming for is at most A3 sized (approx. 11 x 16 inches) prints from my 35mm negatives?

To start answering my own question, and based on memory..

I seem to remember the magic number '300dpi' as a target for a good quality prints.

If that's the case (please correct me if not)

Let's say I'm scanning the full 35mm frame at 4000dpi. If that happens I'm getting approximately (36mm * 4000/25.4 ≈ 5600, and 24mm * 4000/25.4 ≈ 3780) so 5600x3780 pixels.

So then the max print size would be

• 5600 / 300 ~ 18.6 inches (long side)
• 3780 / 300 ~ 12.6 inches (short side)

Which translates to ~32cm x 47cm so a little more than my 'A3 target' above.

I wonder how viewing distance factors into the above though.

Three replies. [edit: you started answering your own question while I was typing]

1 - A quick calculation you have certainly done yourself. A3 is 12 inches on the short side. At the consensus resolution of 300dpi, that is 3600 pixels. Your 35mm frame is ~1inch on the short side, so you need to scan at ... 3600dpi.

2 - Use your eyes, or those of a good-natured guinea pig. Small test prints; change resolutions, (double-??)blind test. I would say, with A3 prints, the subject is not allowed to stick his/her nose on the print.

3 - "Resolution" in scanner specs or tests usually means the highest lpi (2x, actually) where the stripes are still discernible. In technical terms, this means scanner resolution is defined at vanishing MTF. Which is not very meaningful. Below some MTF plots from tests of my V700 10 years ago. Not in the 4000dpi ball park, but just to illustrate the principle.

Straight from scanner:


With the "best" parameters for MTF restoration using Gaussian USM:



In both cases, the MTF drops to zero just beyond 0.4 cycle/pixel. But the boost in MTF at 0.3 cycle/pixel, from ~0.17 to ~0.55, makes a large difference in perceived (sharpness, resolution, whatever, I'm not here for semantics).

Please note that the kind of MTF restoration I'm proposing is much less than what I sometimes see online or in shows, that just hurt the eyes. On occasion, I've brought this up to the photographer, that either admits sheepishly having sharpened maybe a little too much, or takes a stand and claims that's how he likes his pics.

@bernard_L thanks for the insightful reply.

Out of interest, what is the reason for the dip at .6 cycle/pixel in the sharpened example?

albireo said:

@bernard_L thanks for the insightful reply.

Out of interest, what is the reason for the dip at .6 cycle/pixel in the sharpened example?

Click to expand...

1. Spatial frequencies beyond 0.5 cycle/pixel, in actual use, "do not exist". More precisely, they are aliased onto the frequency that is symmetrical wrt 0.5 cycle/pixel. Remember grain aliasing, buzz word occasionally found on internet discussions.
2. How then will you ask, can the plots display frequencies beyond 0.5 cycle/pixel. Because of the trick of slant edge MTF measurement, not applicable in general image scanning. Google it...
3. The actual V700 hardware+firmware is a bit complex (staggered rows of sensors) and must be treated as a black box. Short answer: I don't know.

albireo said:

Or - looking at it from the other side - is it worth unpacking and setting up the 5400dpi scanner if all I'm aiming for is at most A3 sized (approx. 11 x 16 inches) prints from my 35mm negatives?

Click to expand...

From my viewpoint the 4000dpi (if that is its true resolution) will be sufficient to at least 20 inches. Maybe 24. Since you need less than that I wouldn't set up the 5000dpi scanner for this purpose.

5000dpi will be sufficient to at least 25 inches. Maybe 30.

Just my personal experience and taste. Yours may vary.

This is why I built my darkroom. My brain in retirement has no desire to deal with all this technical complication.

albireo said:

is it worth unpacking and setting up the 5400dpi scanner if all I'm aiming for is at most A3 sized (approx. 11 x 16 inches) prints from my 35mm negatives?

Click to expand...

You can only find out by giving it a try. Very simply put, "not all pixels are created equal", and there's no meaningful way to compare at a theoretical level the 5000dpi from one scanner to another's 4000dpi. Make some test scans from various negatives/slides and have them printed.

The proof of the pudding is in the printing.

PS: note that post-processing of scans is a major factor and I'm sure you're aware that the ideal post processing parameters (including particularly sharpening) will differ from one scan to another, often even at the level of individual images.

koraks said:

You can only find out by giving it a try. Very simply put, "not all pixels are created equal", and there's no meaningful way to compare at a theoretical level the 5000dpi from one scanner to another's 4000dpi.

Click to expand...

Thanks Koraks. I'm not seeing all this variability across the two scanners in question (Nikon Coolscan 8000ED and Minolta Scan Elite 5400). At standard viewing sizes on my monitor, and at small printing sizes (less than A4 size) using black and white 35mm film, I am not able to pick any differences between the two given two scans of the same negative and equal (standardised) inversion routines. My post was about finding when/if this equivalence starts to break apart at certain larger sizes.

Perhaps what you say is valid when one consider the broader range of scanning options people use (flatbeds, DSLRs etc) and in that case I might agree with your 'not all pixels are pixels', but in my setup, as I've repeated many times, negative exposure and development variables have IME a far bigger impact on the image than which of those two dedicated film scanners generated my pixels.

I understand what you say. I still say "the proof of the pudding is in the printing". You can't gauge where the break-even point is at a theoretical basis. Been there, done that; doesn't work. But if you want to believe differently, go ahead and see if you can figure it out. I couldn't and had to actually make prints to see what's going on on paper.

To clarify (or make matters more confusing, YMMV): the simplistic approach is to argue that 300dpi is all it takes to render the finest possible detail visible to the naked eye. For this reason, many printing systems are limited to something along these lines (e.g. 240-360dpi). So in theory, your math in #2 would pan out. Then again, in practice, you'll find you can print bigger or smaller than that depending on post-processing, printing process and what you look for in the final print. These factors make the math so fuzzy that it's virtually impossible to tell whether something will look "good" or not. It's complex and to an extent highly personal, and to make matters even worse, it also depends on the specific image. Some images require high acutance as well as high resolving power, while others can convince more than well enough despite being a little fuzzy.

Personally, I find that the limit for a really good print from a 35mm is around A4 size. Anything larger starts to crumble IMO (also optical prints). But I know that plenty of people opine differently and are perfectly happy with far bigger prints from the same size negative. I apparently look for something else in my prints and that limits the size. So whatever answer you'll arrive at will likely be different from mine. All the more reason to test for yourself, because in the end, we can all assert that you will be able to print just fine at X or Y size, but that doesn't say anything about whether you'll be satisfied with the print.

koraks said:

I understand what you say. I still say "the proof of the pudding is in the printing". You can't gauge where the break-even point is at a theoretical basis. Been there, done that; doesn't work. But if you want to believe differently, go ahead and see if you can figure it out. I couldn't and had to actually make prints to see what's going on on paper.

Click to expand...

Well no definitely, I think the final proof is to scan the same X negs with the two devices, blow the prints larger than the 'expected limit' and see if a decision can be made. I will try that. Thanks!

Absolutely, give it a try. And if you can, try a couple of different images. Color & B&W (assuming you also do color; IDK), but also scenes that rely on a lot of fine detail rendering, crisp edges vs. more 'organic' shapes and transitions that work differently, visually.

koraks said:

Absolutely, give it a try. And if you can, try a couple of different images. Color & B&W (assuming you also do color; IDK), but also scenes that rely on a lot of fine detail rendering, crisp edges vs. more 'organic' shapes and transitions that work differently, visually.

Click to expand...

Thanks Koraks - for now I'm concentrating on black and white. I don't think I have a good setup in place for a good, reliable colour workflow.