8mm / Super 8 / 16mm - Scanning Resolution

This discussion has been around multiple times, and the purpose of creating a topic is to, hopefully, compare perspectives.

I would start by admitting that this is the very reason I was drawn to build a DIY scanner, to see what I could get from those 50 year old film reels.
Many have pointed out the limitation of the camera lens, which are very valid points. Nevertheless, the purpose of this community is to scan film, so I am going to borrow this line from @friolator, which I have taken as a mission statement:

to digitally reproduce the film as closely as we can, not so much the picture the film contains

With that, I would like to share a simple perspective of why, in my view, 8mm and Super 8 should be scanned at a higher resolution than typical HD (1080 vertical).

Test Specs - lines/mm
The test film for 8mm (below) calls for lines per millimeter, and the range is from 20 to 60. But if one looks closely at the right side of the chart, the indication references center segments 60-240.

The SMPTE recommended practice for the above is RP19-1982 in item 3.4 indicates

The rosette in the center shall indicate a measurement from 60 to 240 lines per millimeter.

Below a compilation of the dimensions in mm.

If the film under this particular conditions is capable of resolving 240 lines per mm, this is a relevant data point. But many references are expressed in line pairs per mm (lp/mm), or cycles per mm (cycles/mm).

I could not find a definition of the SMPTE lines/mm so I would assume 240 implies resolving 120 white lines equally spaced with 120 background lines, which would then be equivalent to 120 lp/mm (line pairs/mm). That would be the lowest too.

Film Specs - cycles/mm
Film specification for Kodachrome expressed the resolution as Modulation-Transfer Curve in Response percentage vs Spatial Frequency. There is great information on the subject, this example offers great detailed information.

For this discussion, the reference would be the typical Modulation-Transfer Curve of Kodachrome.
MTF Kodachrome

The curve indicates that after 80 cycles/mm the film response is less than 10%, with Daylight exposure, very representative of a typical 8mm film.

Which one is it?
Both. On one hand, the test pattern clearly indicates that projectors (and now our scanners) should be tested to a resolution of 240 lines/mm (here assumed as 120 lp/mm) and on the other hand exposed film is capable of transfer 80 cycles/mm.

Cycles are not Line pairs, but the film space that these occupy, is actually the same.

100 lines,100 pixels?
If in one millimeter one wishes to represent 100 lines (50 white, 50 black), it is representing 50 line pairs, and it can also represent 50 cycles.

So why not say that the digital representation of these are 100 pixels? the more than… in Nyquist.

Nyquist’s theorem states that a periodic signal must be sampled at more than twice the highest frequency component of the signal

To represent 50 cycles/mm one would require sampling at more than 100 cycles/mm, more than 100 pixels/mm. As explained in the example reference, line pairs have even higher frequency components than cycles. Lines are more like a square wave, cycles more line a sine wave. How sharp does one line transitions would be directly affected by the maximum component sampled.

Super 8
If one goes by the film resolution limit of 80 cycles/mm, it would require more than 160 pixels/mm. That would represent a frame of 926 x 667 pixels, and the entire film width/height would require 1264 x 677 pixels.
If one goes by the test standard resolution of 240 lines or 120 lp/mm, it would require more than 1390 x 1000 pixels for the frame. In this case if the entire film width/height is scanned, that would require more than 1896 x 1015 pixels.

Summary
Scanning a full width height Super 8 film at 1920x1080 is barely the Nyquist minimum requirement, and probably under the 240 lines provided by the SMPTE test film recommended practice, yet is probably slightly higher than the film resolution.

Is the bare minimum good enough? Most likely not. Digital processing would yield better results if the source material is a higher resolution, even if the target output is 1920 x 1080.

Is 4K too much for 8mm Super8? If one is doing full width/height, 4K is about 2 times Nyquist. Not unreasonable or excessive.

What about 16mm? If HD is the bare minimum for full width/height on 8mm, it is easy to extrapolate that it would fall short as 16 mm target output.

There are other considerations, such as grain, and even scratches. Circling back to the mission, if the goal is to digitally reproduce the film as closely as one can, (grain, scratches and everything else) 4K is suitable as scanning resolution for 8mm and Super8. Add to that consideration that some cameras/sensors spatial resolutions would be further reduced by the bayer filter.

In closing, one can use the resolution that is appropriate to the content and budget, but keep in mind that 4K for digitally preserving is a reasonable resolution for 8 and Super 8. Here is a sample of 8mm scanned at higher resolution than 4K. Other than levels, cropping, and sizing no other postprocessing performed. Be sure to select 4K playback.

Hope this is helpful and look forward to reading other perspectives/results on the subject.

6 Likes

Supporting this: if you are aiming at preserving the “film look”, including grain, dirt and tiny scratches, and if you can afford to handle the corresponding increase in processing time and storage space, go for the highest resolution you can handle. This will also give you more latitude for possible newer (upcoming) ways to post process your files. One case in point is the upcoming availability of HDR-formats and -devices, which most film scanners, distribution channels and end-used devices currently have a hard time to handle at this point in time.

Even if you do not aim at preserving the original film look, the scan you are doing of a certain film today might be the only scan possible, as certain film emulsions have a severe tendency of color-fading. That is, the scan you do today might not be reproducible in the near future. Again, a reason to go higher with scanning resolutions.

Some other remarks: the pure image content of a Super-8 film is substantially less than what might be appropriate as the scanning resolution. From the limited experience I have with old film stock, I think a good average estimate of the image resolution of Super-8 film could be around 720p. The example you posted above shows, if watched in 4k, scratches and dirt in all their glory. But the actual image stays fuzzy; any picture element I am seeing is stretched out over several pixels. Clearly, the films original image content is over-sampled to a noticeable degree. That you do see a difference between watching this footage in HD or 4k is probably related more to the variances of the compression algorithms at work at different resolutions.

Note also that the SMPTE-test film shown above most probably was not produced with standard film stock. Most certainly, a higher resolution film stock was used for this. As your MTF plot of Kodachrome stock above shows, even 100 cycles/mm were out of reach for one of the sharpest films that consumer could buy in the old days.

Summarizing: one should scan (for archival purposes) with the highest resolution which is deemed to be affordable. For distribution channels like YouTube or so, it might be also be advantageous to go to the highest resolution possible, just to preserve the existing image definition. For other distribution channels, much lower resolutions might be possible, as the native resolution of Super-8 footage (combined camera and film resolution limits applied) is most probably less than HD.

Thank you @cpixip well said. I agree with many of the points, and would highlight this one in particular.

I would also add, that your extraordinary work in postprocessing validates trade-off between resolution and multi-exposure/bit depth for the resulting output file.

This is a valid point. The Recommended Practice description:

The test film shall be produced as a camera original film photographed on high-contrast, high-definition, positive-type motion-picture stock made in accordance with American National Standard for 16-mm Motion-Picture Film Perforated 8-mm Type R, 2R, ANSI PH22.17-1982.

Not sure how available was, but high contrast film like Eastman 5363 for example, would have supported the additional resolution. Below from the datasheet:
MTF Kodak 5369

70% response for 120 cycles/mm, approximating for 120 lp/mm = 240 SMPTE lines/mm.

Black&White
I missed it completely in the post above. There is a lot B&W original film, it is worth adding it to the resolution consideration.
MTF Kodak 7302 BW

Kodak 7302 in this case, 10% response is about 150 cycles/mm, almost double the response of Kodachrome.

Using this resolution with the prior pixel calculations for Nyquist, 2370 x 1269 for the full width-height of Super 8.

B&W is also abundant in 16mm camera original film, and at least from the film perspective, higher than 4K may be justified. Always with the caveat of limitations other than the film itself, such as the quality of lens used in the exposure.

2 Likes

I can only make a small experimental contribution to this subject.As I use a Pi Camera HQ I compared visually on a super8 frame a very enlarged level of detail in mode 3 full resolution 4056x3040 and mode 2 binned 2028x1520, for me no doubt the higher resolution brings a little something.

I would have liked to do my capture in mode 3 but it is really too slow. In the end, to have a decent fps I capture directly in 1500x1080 without any border. I get a good result but I have some regrets not to be able to use the full resolution of the HQ.

1 Like

Just to continue the discussion slightly, I want to present a little bit of the variation of image definition I am seeing in my scans.

Probably the worst footage I have encountered so far is from a 1975 Agfachrome movie, recorded with my own Chinon-manufactured Super-8 camera. This one was scanned using a see3cam-unit at the original resolution of 2880 x 2160 pixel, for speed reasons transferred as .jpg. The lens used in this scan is the trusted Schneider Componon-S 50mm.

Focus was on the right bow of the brigde. I selected this frame on purpose, because of the dirt particle in the left part of the frame - this allows one to judge the quality/focus of the scan.

Clearly, this frame is far from coming even close to HD resolution.

Compare this to the following frame:

This footage was recorded about 10 years later, using the Moviechrome 40 film stock from Agfa, which replaced the old Agfachrome stock in the beginning of the 80’s. Coming from an Ebay-find, the camera used is unknown. But it is clearly better than my old Chinon.

Scanning was done this time at a resolution of 2028 x 1520 pixel with a Raspberry Pi HQ camera, fitted with the same lens as above. The camera was operated in mode 3.

This frame comes close to a HD-resolution. Again, this frame was specifically selected to have some dirt and scratches in order to judge the raw image definition of the scan.

For comparision, here’s finally a scan of Kodachrome 40 footage, also from the 80’s. The scanning setup is the same as before (Raspberry Pi HQ camera with Schneider lens). The movie camera used is again unknown, as this comes from an Ebay-find as well.

There is a small dirt particle visible on the right collar of the girl (in case you want to judge the scan quality). In terms of image definition, the Agfa Moviechrome 40 and the Kodachrome 40 seem for me to be comparable. Note that all examples were chosen from footage which did not include much camera movements. But in neither of the footages a tripod was used.

Additional comment: As it seems that the forum software recodes uploaded images and scales them to a maximum width of 1920, here are some enlarged parts of the above images. First, the Agfachrome film stock digitized with a see3cam (300% enlargement):

If one looks closely, .jpg-compression artifacts are visible. Nevertheless, the film stock simply does not have that good of a resolution. As far as I know, there is no MFT-data available for that old material, as measuring MFTs became only fashionable in the 80’s (and even then, they were not easily available).

Next is the Moviechrome 40 film stock, also from Agfa, which was introduced at the beginning of the 80’s. Here, the scale-up is again to 300%. So far as I know, there is not MTF-data available for this film stock either.

The above image was captured with a Raspberry Pi HQ camera running in mode 3. Finally, here’s an example scan with the same setup, this time Kodachrome 40 film stock

For Kodachrome, MFT-data is available (as shown in the first post above).

4 Likes

Does anyone have a PDF copy of the paper “The Image Resolution of 35mm Cinema Film in Theatrical Presentation” from the SMPTE Imaging Journal, vol. 113, No. 2&3, Feb./Mar. 2004?

Here’s a mention of it:

In the early days of High Definition production, many in the motion picture and television industries were reluctant to give up their tried and true method of capturing the images on 35mm film. Hank’s research (1985 -2003), and seminal 2003 paper on “The Image Resolution of 35mm Cinema Film in Theatrical Presentation” shook up the Hollywood production community. Hank’s research and testing was part of the ITU-R Study Group 6 work related to Large Screen Digital Imagery (LSDI). A test was performed to determine the end-to-end resolution provided by 35mm cinema film projection in typical motion picture theater. A resolution test pattern was photographed on a 35mm negative film stock typical of those used for the production of feature films, and 35mm inter-positive, inter-negative, release prints and answer prints were struck from that negative in conformity with cinema routines.

This research provided definitive proof that the 1920×1080 High Definition video format could deliver comparable or better quality to that of 35mm film in the cinema and ultimately to the consumer’s home.

Today, the vast majority of motion picture and television production is acquired electronically in High Definition or UHDTV.

It is part of the IEEE library, this is a link to it

PS. And this is the datasheet of the Kodak type 5274 used for the test.

If anyone has some 8mm SMPTE test film I’d like to borrow some.

It’s a bit more involved than this. If you’re scanning with a CFA camera you get at best 3/4 true resolution when you debayer, and more realistically about 2/3rds compared to doing a monochrome capture. Your camera will have an optical low-pass filter as well to reduce aliasing/Moire, this blurs the image and further reduces the true resolution of your capture. So your true resolution is probably around 50-65% of the native sensor resolution. This is why black-and-white film should really be scanned with a monochrome camera, you will get a much better capture. So what this means is if your sensor is 2K like comes with the current model Retroscan (2048x1536) then the true resolution is more like 1024x768 - 1364x1024 which is lower resolution than 1080p in the most generous and best-case scenario.

If you were to use the monochrome version of the same camera though things improve substantially - now there’s no debayering, just the low-pass filter, so you would perhaps get 80% or so true resolution. That’s just spit-balling a number you would have test the camera with and without the low pass filter and use SMPTE test film to really know for certain what you have. 80% is 1638x1229 which is indeed higher resolution than 1080p.

Not all cameras have an optical low pass filter. The DSLR camera I use D3200 does, the next one, the D3300 doesn’t. I also could not find a low pass filter in the Raspberry HQ (IMX477), only the IR filter with the instructions to remove it.
On the subject of monochrome sensors, there is this website offering to remove the sensor bayer, including the Raspberry HQ. This website offers similar conversions for Sony cameras.

You want one for scanning though. You would not want a camera sans-LPF for film scanning if you have the choice.

Not sure the reasons, specially when the discussion is in the context of using higher resolution sensors for better sampling rate. The opposite (anti-aliasing filter for lower resolution sensors) is understandable.

Nevertheless, I found this interesting reference on the RpiHQ sensor, and it appears the filtering is performed electronically.

I used to scan film using a RED Epic Dragon 6K and discussed this subject with the main color science guy at RED. This camera system has a removable OLPF module and it was knocking off resolution when scanning film.

…so I bought a used OLPF module on ebay and replaced the OLPF glass with custom ordered optical clear glass from Midwest Optical Systems, Inc

There was zero aliasing and my scans were sharper. Film images don’t have fixed sampling pattern issues because film grain is random and changes all the times. I scanned a LOT of film with this system haven’t seen a bit of aliasing - not even on perf sides.

There’s an IR cut filter in front of sensor of my 5.3K Flir Oryx and will ask if there’s OLPF properties in the glass - I highly doubt it.

2 Likes

Confirmed by Flir that there are no OLPF In the Oryx. Only IR Cut that can be removed but it voids warranty. Mono Oryx have optical clear for sensor protection.

…and if it did it would be detrimental to film scanning.

1 Like