Film Photography & being mindful of Scanning
(Note: I am giving Medium a try as I find the layouts look nice. It's here)
Film often gets the short sell. Digital came onto the scene and it no longer made sense for the vast majority of photographers and snapshot shooters to keep buying rolls. But often I observe digital shooters taking cheap shots at the medium. They mistake soft scans for a soft medium. And while the nature of film inherently lacks the sterile clarity of digital, much of that softness comes from a different source.
Which is why I wanted to attempt a visualization the impact scanning can have.
But before that comparison we need to understand that the resolution of a camera or a scanner are not all created equal. The Resolution in MP(megapixels) or dpi(dots per inch) listed on the box often isn’t a literal translation to what is resolved (the fine detail we can see).
Shooting digital it’s very easy to take this distinction for granted but it still applies. Am image from a 24MP camera might only be resolving 15MP worth of detail information. Dxomark is actually a great resource to understand how a sensor lens combination might not perform as expected.
Which begs the question on how to express this difference on what you could get vs what you do get. You could use megapixels to reflect what’s resolved. Dxomark tries to do this with their Perceptual Megapixel but it invites confusion. There are MTF charts or other graphic illustrations. They are more accurate but a bit difficult to decipher. In my example I settled on the line pair because its ease as a visual metaphor.
The basic element of a line pair is simply a dark and light line next to each other. Which can then be grouped over a distance to represent some measure of detail. If the transitions between light and dark lines can still be seen then that many line pairs over that unit of distance are resolved. Typically either picture height (lpph) or mm (lpmm) are used.
Adjacent is an example of 5 line pairs over a distance of 1 millimeter. There are a total of 10 lines (5 black and 5 white) which would need at least 10 pixels to be able to represent in a final image.
It’s not the perfect metric as the real world is rarely filled with such high contrast and regular details, but it should be easy to imagine how it relates to images beyond black and white bars. For example something that might be a good analogue are windows on a skyscrapers from the header.
 |
| Scale represents 1 mm of the film |
These 7 transitions are happening over approx 0.5 mm on the actual negative. This gives us 14 lp/mm. So details at this frequency are being resolved.
For fun we can try scaling to the smallest size that can reproduce those details. One way to test this is to reduce the image so the resolution matches our 14 lpmm estimation of those windows. Any greater of a reduction and the lines start to blend together and we lose our ability to discern the windows. To prove that we did this right we can observe a false patterns as the scaling algorithm tries to place these alternating lines into a set of pixels that doesn’t quite fit.
 |
| 125px | 135px |
But we can add a little bit of further nuance. For example we can apply a blur filter (far left image) and sharpening filter (far right image). The image resolves the same windows but with different contrast.
 |
| Blurred |Normal| Sharpend |
Which brings us back to scanners and how they might impact how we see film. If a scanner isn’t able to resolve a specific pair of lines then it’s acting as a limit to the detail we might have from our film. But beyond that even if a scanner can resolve the lines it might resolve them softer then another scanner.
From some consumer scanners I’ve had access too I wanted to start with the nominal box resolutions:
- FUJI (Lab standard) — 1200 DPI — (1728 * 1152) — 1.99mp*
- Nikon CS 8000–4000 DPI — (5616 * 3744) — 21.03mp
- Minolta 5400–5400 DPI — (7632 * 5088) — 38.83mp
- Epson V700–6400 DPI — (9072 * 6048) — 54.87mp
- Plustek 8100–7200 DPI — (10224 * 6816) — 69.70mp
These scanners reach an impressive 70mp clearly beating all current 35mm consumer cameras. But to utilize those figures the scanner would need to be able to tell apart lines near that resolution, and the film negative would need to contain at least 70mp worth of detail.
So to start each scanned a 1951 Resolution Chart. It might not be the best target, but it matches very well with our idealized line pair from before. The top images are an vector version made to match the scanners results.
 |
| Top: Ideal - Bottom: Measured |
These are 100% center crops. Notice that the highest resolution scanner didn’t produce the highest resolution result. In my opinion of multiple runs the breakdown happens as follows.
- FUJI (Lab) — ??? (Not Tested)
- Epson V700 — 2580~2896 (dpi) — 8.9~11.2(mp)
- Nikon CS 8000 — 3251~3649(dpi) — 14.1~17.8(mp)
- Plustek 8100 — 3649~4096 (dpi) — 17.8~22.5(mp)
- Minolta 5400 — 4598~5161(dpi) — 28.3~35.7(mp
But beyond what lines we can make out is how they become gray as they shrink in size. If we had perfect reproduction we would look much closer to the reference images above each frame. Even with a sharp target the results are imperfect. But it’s not as bad as the above might suggest. Film has it’s own losses and the differences wouldn’t be so drastic in a real image.
 |
| Ektar 100 |
Each photo was scanned multiple times at the max optical resolution. Because each image was a different size and any comparisons should target a final output size the results where then scaled for a reasonable 300dpi@8x12 print. As this isn’t intended to be a lens test a digital photo was taken with the same lens to eliminate it as a major variable.
 |
| Nikon, Epson, Minolta , Plustek , Digital |
As you compare between each result it’s pretty clear that a lot of the subtle details are lost on the flatbed (notice the small lines below the ladder). The plustek and Nikon are close in a way the 1951 target would not have suggested. And finally a bit better then the crowd the Minolta pulls ahead.
As each scanner produces a bit more detail, none of them really jump out as far ahead as the 1951 target might imply possible. It might be tempting to suspect the Minolta has exceeded the limits of the film used. The scan clearly is not filled with 35mp worth of detail (Well maybe grain detail). So what is the take away?
It’s managed expectations. It’s understanding that for those who shoot film in a digital workflow we depend on the tools that bring them into the computer. It’s not the we should blame scanners, we should just realize that what we see on the screen is as much a reflection of the scanner we choose as the film we shot.
But how far is it worth to go?
Of course some will be curious as to where the limit is. I can’t provide an answer, but it is possible to to estimate how far we could go. We would need to start with how far can we take film? There are plenty of resources on films absolute resolution. Zeiss has made claims to films max resolution in CLN17 CLN19. They cite very generous figures of 200lpmm but without examples. California Creeks lists the more realistic 1.6:1 contrast figures alongside the fantastical 1000:1 figures. We see 80lpmm is a more realistic limit. The best attempt to provide realistic figures lens included was one done by Tim Vitale suggesting ~60lpmm color and ~80lpmm for BW. Clark vision is a tad more pessimistic and puts it closer to 55lp/mm for the limit.
I’m not going to disagree with these figures above. If your explications are from what you expect from digital it’s going to be even lower. But it isn’t all that bad for films future with resolution.
We can move away from a scanners as a good enlargement always has a chance to improve results. Tim Parkin did a various number of comparisons and found even with 35mm color film it wasn’t as bad as the other authors suggested (Even including images!)
And this comparison (optically printed) speaks to the argument that resolution war isn’t completely won (If you are willing to use slow bw micro films)
The gold standard of Drum scanning is noted for being both able to reduce the presence of grain artifacts while reaching astronomical resolutions approaching nearly 190mp for a 35mm frame. While impressive, that extra resolution has a diminishing benefit. (If you are interested in such a scan you can view one here)
But to put my own opinion in writing. The low contrast figures on a particular film seem about the realistic upper estimation on what you can expect, and about half the scanners optical resolution is about what you can get (Assuming that number doesn't exceed the estimation).
Of course want to see a full resolution result to pixel peep a 100% crop is below. Again the same lens was used. No additional sharpening was applied, and minimal post to match color.
 |
| V700 | Minolta | Digital |
Links and Mentioned Sources in order:
- http://www.dxomark.com/Reviews/Looking-for-new-photo-gear-DxOMark-s-Perceptual-Megapixel-can-help-you
- https://en.wikipedia.org/wiki/1951_USAF_resolution_test_chart
- http://www.zeiss.com/content/dam/Photography/new/pdf/en/cln_archiv/cln17_en_web.pdf
- http://www.zeiss.com/content/dam/Photography/new/pdf/en/cln_archiv/cln19_en_web.pdf
- http://cacreeks.com/films.htm
- http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2007-04-vitale-filmgrain_resolution.pdf
- http://www.clarkvision.com/articles/film.vs.digital.1/index.html
- http://www.trenholm.org/hmmerk/scanelite.html
- https://www.onlandscape.co.uk/2014/12/36-megapixels-vs-6x7-velvia/
- http://www.imx.nl/photo/leica/camera/Leica-camera/styled-34/
- http://empiricalnon.com/rab.html
- http://www.karimsahai.com/wp-content/uploads/docs/Screen_SG-8060P_Mark_II_Drum_Scanner_karimsahai.pdf
- http://plustek.com/plustekdoc/How%20to%20apply%20a%20proper%20scanning%20solution%20on%20your%20film%20scanner.pdf
Comments