How Many Pixels? for Medium Format?

ColSebastianMoran

ColSebastianMoran

( IRL Richard Karash )
Local time
11:17 AM
Joined
Sep 15, 2010
Messages
2,738
How many pixels? My interest is "How many for excellent results for a specific use, e.g. a print." (My largest prints are 20x30".)

I know from other tests that I don't get much or any additional detail from my 35mm films beyond 24Mpx. But, how about 120?

This experiment is with a very good 120 negative, 6x9, from Mamiya Press with the excellent Mamiya 100 f/2.8 lens on good color-neg film. Subject has good detail, shot in very good light. Scanned multiple ways, files from 12MPx (flatbed) to 24MPx (cam-scan) to 100MPx (stitched). Conversions in NLP. Pixel-peeping, I can see differences in the files. More detail resolved, better feeling of the materials and surfaces. So, to "get it all" or for archival purposes, I want the larger, stitched file for medium format.

But, will I see differences when printed? So, I make 20x30" prints of each file. The results surprise me:
- Very hard to see a difference 100 vs 24MPx at this print size
- Easy to see that both are better than print from the 12MPx flatbed file
- But, even print from the flatbed file looks pretty good and would be usable

My conclusions:
- For archival, to "get it all," shoot or stitch 16MPx per square inch of negative
- 24MPx is enough for good prints (e.g. my 20x30's) and all my uses
- I will shoot everything at 24MPx, but save the negatives for potential re-scan
- Perhaps stitch only a very few special images
- Hypothesis: Dealing with film flatness may be more significant for good results today than adding lots of additional MPx.

Here are two prints, 100 and 24MPx for illustration only, small area of the full print, iPhone shot. Just to show one of the areas in which I find it very hard to see differences. (This for illustration, to look for differences yourself, I think the only way is to inspect the files or make prints; see links below.)

201011-Cheers-Comparison-100-vs-24MPx-IMG_7656-ED-1kScr.jpg


Here is the full image:

200917-CheersPub-24MPx-DSC1546-1kScr.jpg


For anyone who would like to look at the files in full resolution:
- 24MPx one-shot: http://2under.net/images/200917-Cheers-24MPx-DSC1546-MaxSh.jpg (18MB)
- 100MPx stitch: http://2under.net/images/200917-Cheers-100MPx-DSC1551-59-MaxSh.jpg (85MB)

Comments invited! My best regards to all... I have learned so much from all of you here on RFF.
 
On my old screen I do not see a difference. I have a 36mp ff DSLR that has pixel shift and I can hardly deal with the files. Doing pixel shift correctly I definitely see a difference but I'm not sure if it would show on a print. But then you say you can see a difference with the negative stitch so I guess the huge file is worth it for a very large print.
 
On my screen, if I zoom in I get jaggies with the 24mp image but not the 100mp. It doesn't look like 100mp is making a significant difference in that it looks more like it's anti-aliasing than actually resolving more detail to me.

There might be some scanning-related artefacts with the 24mp image, the edge of the blue sign where it comes out from behind the lamp-post seems to have a pale edge on the 24mp more pronounced than on the 100mp image.

All of which might just mean that the 100mp is resolving more detail, but the film doesn't really have any to offer. At any sensible viewing distance they are going to look indistinguishable.
 
joe bosak said:
There might be some scanning-related artefacts with the 24mp image, the edge of the blue sign where it comes out from behind the lamp-post seems to have a pale edge on the 24mp more pronounced than on the 100mp image.
Click to expand...

Yeah, that looks like a sharpening artifact. I used Lightroom Export... Sharpen for Glossy Paper... Standard for both files. The artifact is proportionately bigger in the smaller file. I can see it on screen pixel peeping, but can't see it naked eye on the prints.
 
You all are brave for inspecting the files. Looking in LR, before output sharpening, I do see more detail and the surfaces/materials look somewhat more real when pixel peeping.

But, seriously, I have real trouble seeing any differences in the 20x30" prints.
 
Update. I printed all the files at 20x30"

Here are my conclusions:
- Very hard with naked eye to see any difference between the prints from one-shot 24MPx vs stitch for 100MPx. Very hard. Yes, with a magnifying glass, just as we see differences via pixel peeping.
- Surprise: Same 6x9 negative scanned with a flatbed produces a "not too bad" 20x30" print. In the past, I've limited my prints from V500/V600 to 12x18. For a non-picky viewer, the print from flatbed would pass, and that surprises me. OTOH, I'm a pretty picky viewer.
- I don't see myself wanted bigger than 20x30" prints, so I'll stick with one-shot, good lens, 24MPx. For a few images, I'll stitch, not for the print, but for archival purpose.
 
Excellent! I somewhat suspected the result as I never make that size prints. I've always been satisfied with my flatbed scans of 6x9 negatives (scanner similar to yours).

I will tell you my 36mp ff camera with pixel shift (which I don't really understand) and processed through RAWtherapee (using their pixel shift modules) are really amazing. But it is a hassle and I might as well just use my Mamiya 6x9.
 
charjohncarter said:
...I will tell you my 36mp ff camera with pixel shift (which I don't really understand) and processed through RAWtherapee (using their pixel shift modules) are really amazing. But it is a hassle and I might as well just use my Mamiya 6x9.
Click to expand...
With the Bayer array of a typical sensor, each pixel is a merger of data from 4 photosites; green, green, red, blue. If the sensor could be persuaded to produce all colors from each photosite you’d have 4x the detail, less moire, and less chroma noise.

That’s the idea behind Pentax’s “pixel shift” function. It’s done by taking 4 exposures, with the sensor moving the distance between photosites for each exposure, in a box pattern like right, down, left, up, or some equivalent pattern. Then the four exposures are merged in software. But the camera would have to be rigidly mounted so as not to move between exposures.

That last part was always true for the original Pentax K-1 but there’s another option for the Mark II version of the camera, which is “smart” enough to cope with hand-held pixel shift exposures. But this Dynamic Pixelshift is done a bit differently and the result is not quite as good as the original method; a tradeoff for convenience.

Sony has a similar feature, and so do Panasonic and Olympus but using 8 exposures while the sensor shifts a full pixel distance.
 
Somehow, the 24MP on the right looks sharper to me on this 15" Macbook Pro.

Edit: And, on my 21.5" iMac as well. I'm looking at the bricks and the venetian blinds.
 
Doug said:
With the Bayer array of a typical sensor, each pixel is a merger of data from 4 photosites; green, green, red, blue. If the sensor could be persuaded to produce all colors from each photosite you’d have 4x the detail, less moire, and less chroma noise.

That’s the idea behind Pentax’s “pixel shift” function. It’s done by taking 4 exposures, with the sensor moving the distance between photosites for each exposure, in a box pattern like right, down, left, up, or some equivalent pattern. Then the four exposures are merged in software. But the camera would have to be rigidly mounted so as not to move between exposures.

That last part was always true for the original Pentax K-1 but there’s another option for the Mark II version of the camera, which is “smart” enough to cope with hand-held pixel shift exposures. But this Dynamic Pixelshift is done a bit differently and the result is not quite as good as the original method; a tradeoff for convenience.

Sony has a similar feature, and so do Panasonic and Olympus but using 8 exposures while the sensor shifts a full pixel distance.
Click to expand...

This is my Pixel Shift test, first the full frame:

Pentax 35mm f2.0 FA-AL by John Carter, on Flickr

And this is a center blow up:

Pentax 35mm f2.0 FA-AL by John Carter, on Flickr

The original, and the blow-up are at f 2.0 Pentax FA f 2.0 35mm lens.
 
ColSebastianMoran said:
How many pixels? My interest is "How many for excellent results for a specific use, e.g. a print." (My largest prints are 20x30".)

I know from other tests that I don't get much or any additional detail from my 35mm films beyond 24Mpx. But, how about 120?

This experiment is with a very good 120 negative, 6x9, from Mamiya Press with the excellent Mamiya 100 f/2.8 lens on good color-neg film. Subject has good detail, shot in very good light. Scanned multiple ways, files from 12MPx (flatbed) to 24MPx (cam-scan) to 100MPx (stitched). Conversions in NLP. Pixel-peeping, I can see differences in the files. More detail resolved, better feeling of the materials and surfaces. So, to "get it all" or for archival purposes, I want the larger, stitched file for medium format.

But, will I see differences when printed? So, I make 20x30" prints of each file. The results surprise me:
- Very hard to see a difference 100 vs 24MPx at this print size
- Easy to see that both are better than print from the 12MPx flatbed file
- But, even print from the flatbed file looks pretty good and would be usable

My conclusions:
- For archival, to "get it all," shoot or stitch 16MPx per square inch of negative
- 24MPx is enough for good prints (e.g. my 20x30's) and all my uses
- I will shoot everything at 24MPx, but save the negatives for potential re-scan
- Perhaps stitch only a very few special images
- Hypothesis: Dealing with film flatness may be more significant for good results today than adding lots of additional MPx.
...
Click to expand...

In way of adding to the information base:

When I was scanning with the Nikon Super Coolscan 9000 ED, I normally just set the scan resolution to 4000 ppi. With a 6x6 negative, that nets an un-interpolated print size with image area 29.5 x 29.5 inches. I so rarely print to that large a size that it was unimportant to me to keep that huge scanner around.

A 39 Mpixel square crop on an X system Hasselblad (X1D, 907x) 50 mpixel camera nets superbly beautiful 21x21 inch image area @ native 300 ppi output (6200x6200 pixel output).

The key difference from other cameras in capture is that the Hasselblad 50 Mpixel sensor and .3FF/.FFF output files have 16-bit tonal/color resolution (up-rezzed from the sensor's native 14-bit output by the in-camera raw handling signal processor). This makes the 907x outstanding for capturing 6x6 to 6x9 negatives with the native XCD 120 or adapted V system 120 macro lenses ... A full frame capture of a 6x7 negative (to match the native sensor format) nets native 300 ppi image dimensions of 27.6x20.7 inch size with incredible editability.

That file can be uprezzed by sqroot2 on each dimension to net superbly beautiful 39x29 inch print area at 300 ppi output resolution (the resulting file is ~102 Mpixel in size, or double the original size output by the camera). Such extraordinary pixel resolution is only really important for the images most critical on resolving exceptional detail, but it's nice to have that kind of capability if you need it. It's also beyond the capabilities of all but the most well-focused/exposed/processed film images. ;)

G
 
After exhaustive tests back in 2013 to find a definitive answer to the "how many pixels" question, I came up with one: a 36 MP full-frame digital camera = 645 film. This was tested under the following conditions to (a) maximise differences between the two image types whilst (b) minimising sources of error and (c) minimising degradation of image quality between taking it and printing:

• Digital camera: Nikon D800E (36 MP)
• Film camera: Mamiya 645 (Portra 160 film) + Hasseblad Flextight X1 scanner
• Same lens: Mamiya 80/2.8
• Same postproduction software: Capture One
• Same print process and size: A1 (professional lab inkjet, A1 size = 23 in. or 58.5 cm height).

Reasons for choices:

• The cameras were those I used regularly
• I was interested in good colour negative film vs a top-end full-frame digital camera
• The lens is exceptionally sharp with few aberrations
• The Flextight scanner is the best (blows Nikons into the weeds, V700/800 not even close - not surprising considering it cost £12,000 ($16,000), but now sadly discontinued)
• Inkjet prints chosen over C types (though I prefer the latter) because they appear slightly sharper and thus emphasise differences in resolution
• The ends of the Nikon image were cropped to match 645 film (not necessary, but makes it easier to compare prints if they look identical)
• A1 print size is at the limit of print resolution for 36 MP - at that size, about 220 dpi.

I made several different test prints of different images taken in a studio using a tripod, ensuring the film and digital images were as identical as possible, with matching framing, exposure, focus, etc. They were postprocessed identically (though I recall the Mamiya 645 photos needed the exposure lifted a tad to match the Nikon).

(As an aside, I also tested print resolution: the above (220 dpi) vs 300 dpi from the Nikon. I moved the camera nearer a subject, so I could match part of a 220 dpi image with a 300 dpi print, and compared prints of identical crops. The 300 dpi print was definitely a bit sharper when both prints were viewed side by side. But viewed on its own, the 220 dpi print looked sharp. Conclusion: 36 MP can be printed at A1 size but that's pushing it to the limit - ideally, print at A2 (16 in. or 42 cm height) to get 300 dpi. The 300 dpi resolution isn't arbitrary - it's been used for decades, well before computers were invented, and is the approximate limit of human vision.)

Anyway, back to the main point... I compared the Nikon and Mamiya prints myself, and then asked several other people which photo in each pair they thought the sharpest, without telling anyone anything. These included photographers (I was doing my MA Photography degree at the time) and non-photographers. All prints looked equally sharp to me, and that was also the majority opinion. (However, most people preferred the look of the Portra photos because of the colours!)

Lastly, I calculated the resolution of the prints using optics theory, to ascertain what physics says about real-life digital and film resolution. Manufacturers' print and film resolutions are meaningless because they don't apply to the images we see on screen or in print, and don't take into account degradation by the lens/camera, scanning, printing, etc. My calculations came to the same conclusion as the empirical observations above: no difference between the film and digital prints. I posted the calculations on RFF some years ago: https://www.rangefinderforum.com/forums/showthread.php?t=138124

I believe my tests are more rigorous than anything I've seen on the web, which are mostly unscientific and made with poor equipment and judgment.

CONCLUSION: Medium format 645 colour negative film = a 36 MP digital camera for photos viewed/printed at the same height.
 
jzagaja said:
Sigma DP2 Quattro has more details than 6x9 from GW690. Film has nice look.
Click to expand...

That's a curious result. A competently made drum scan of a properly focused 6x9 Fuji would offer far more detail than an APS-C camera. Maybe there was something wrong with the Scanmate?

I'm not familiar with that Sigma but a reported max resolution of 5424 x 3616 is not comparable to that big Fuji at all. Even a 6,000dpi drum scan from the Fuji would produce a file nearly four times the size of the Sigma and with much more detail. (Now whether anyone actually needs that much detail is another question. ;-)
 
I have had 6 drum scanners and this one had smallest register error. I have at flickr Copex Rapid sample but shutter time 1/60 blurred result thus Digibase sample. Sigma was ASA400 to keep short shutter - different light. Field of view:

31350052527_ce5a4b254e_c_d.jpg
 
You must log in or register to reply here.
Back
Top Bottom