Making a 8mm film scanner

[Planefreak]

I was hoping you guys might have some ideas on making a film scanner.  I have over 12 hours of super8 film that I want to scan, but haven't come up with a truly great solution.

I know that the there are a few methods for film scanning, and that for years people have been using the camcorder pointed at a mirrored box or white sheet method, but this requires a good working projector, which I don't have, and this doesn't provide good video.

The two methods that I have seen work well employ either a constant scan method, which feeds the film at a constant rate, then uses a high speed camera to capture each frame as an image.  The other method is to use a slow speed camera, pausing the film, taking a picture then advancing to the next frame.  The pausing method is employed by this machine https://www.amazon.com/Wolverine-Super8-Digitizer-Film2Digital-MovieMaker/dp/B01KA32HH0/ref=sr_1_1?ie=UTF8&qid=1484041178&sr=8-1&keywords=wolverine+8mm+and+super8+reels+movie+digitizer.  I'm not against purchasing it, but it takes away a lot of the freedom and fun of building my own system.

I have 3 or 4 projectors, but none work well.  Many have broken parts, so I've been thinking about taking the drive and feed mechanism from one projector, possibly adding a stepper motor or something with precise control, then using a microcontroller to drive the motor and sync the camera.

The problem is that I'm not really sure how to go about getting a good image.  Anyone know enough about optics/photo recording and what I would need to do here?

For photo recording, I could use my DSLR (prefer not to because this would definitely go over the shutter life expectancy), a GoPro (takes photo/video), camcorder, and older digital camera.

Thanks for any suggestions!

[ajb]

What sort of resolution do you think you can you usefully extract from this 8mm film?  It's probably worth doing a couple of single-frame test shots with the DSLR first to see what you can get.  You may be able to get everything you want out of the film with a much more modest camera, possibly even a webcam, and save the wear and tear on your DSLR.  Then you could do a fairly crude lashup with something like a RasPi controlling a couple of motors to advance the film over a light box that uses OpenCV to line up each frame, capture it, and then shuffles it off to a network share (If my math is right, 1280x1024x24xRGB = ~340GB/hr uncompressed).  That said, a big thing to consider with anything DIY is speed--if it takes 2s to capture a frame, then each hour of 24fps film will take 48hrs to capture.

[grumpydoc]

Have a look at this guy's stuff http://www.super-8.be/s8_Eindex.htm, more stuff at http://www.cine2digits.co.uk/

[jm_araujo]

Some time ago I "hacked" a solution for scanning some old super 8 films from a co-worker.
It was very crude, I used a old projector, replaced the bulb with a old small lcd backlight, removed the shutter and the gate (the film usually is exposed outside the gate area, never before seen footage ), and added a small smd led near the film that lit when the frame was still.

For the camera I used a Point&Shoot with a binocular eyepiece attached in front to get me a zoomed macro field capturing all the image and the led.
Using a slower film rate than half the camera movie frame rate (to ensure every frame was captured), I then extracted all the frames individually (it helped the camera recorded in MJPEG), and using a python script filtered only the ones needed "looking" at the led. Then cropped, assembled it back into a movie file and post-processed for color correction and image stabilization.

To extract and reassemble the frame images I've used ffmpeg. For cropping I used image-magick.
I processed one movie at a time  in a temporary ramdisk, it was way faster dealing with all the frames extracted as files.

The only real quality problem was blowout when transitioning from dark to lit scenes, the camera was slow adjusting the gain, but it was still acceptable.

For all this crudeness the results where good, and my co-worker was quite happy.

Probably you're after a better solution, but I hope the description of my method will help you in some way  .

[Kleinstein]

Taking the picture one by one produces a lot of data. So it take quite some computation to bring back all the data together.

One could still use the DSLR camera - if it has a life view mode so the shutter and mirror are not moved, but using an electronic shutter. As the resolution of the single 8 mm frames is limited, there is no need for a high quality camera. So a better web cam could be good enough.

[helius]

Is your purpose documentary or artistic? The usable resolution on Super8 is probably 720x480 or less, but if you want to capture the film grain effect, Kodachrome edge sharpness, etc you will need a higher resolution.

[helius]

It may be difficult to achieve frame alignment using steppers unless you have optical feedback. Even with a real projector's pull down claw, alignment can be pretty unstable. The classical way to get good alignment is to use pin registration, but there's no projector that does that to my knowledge.
If you can use optical processing, alignment may not matter. The Kinetta uses a continuous feed over a smooth drum (no gate or claw) and reconstructs frame lines digitally.

[grumpydoc]

Is your purpose documentary or artistic? The usable resolution on Super8 is probably 720x480 or less, but if you want to capture the film grain effect, Kodachrome edge sharpness, etc you will need a higher resolution.

About that.

The best film stock with a good camera lens could perhaps reach 100lines/mm, the frame size is just 4mm x just under 6mm so you could make an argument that it is the equivalent of "up to" 720p.

[darrellg]

I was hoping you guys might have some ideas on making a film scanner.  I have over 12 hours of super8 film that I want to scan, but haven't come up with a truly great solution.

Take it to Costco. They have a film scanning service that will transfer it to DVD for you and save you a ton of time. Their prices are reasonable.

[helius]

The best film stock with a good camera lens could perhaps reach 100lines/mm, the frame size is just 4mm x just under 6mm so you could make an argument that it is the equivalent of "up to" 720p.

Yes, no doubt that such resolutions are possible, but Super8 cameras were not usually built with such precision; they were made for home movies. Also, just because a certain MTF limits the resolution of the image, there are very good reasons to use much higher resolution sensors to scan it. It permits noise to be removed before rescaling, producing cleaner images. The optical filtering and contrast reduction is also a major issue, since positive film materials have very high Dmax.

[grumpydoc]

The best film stock with a good camera lens could perhaps reach 100lines/mm, the frame size is just 4mm x just under 6mm so you could make an argument that it is the equivalent of "up to" 720p.

Yes, no doubt that such resolutions are possible, but Super8 cameras were not usually built with such precision; they were made for home movies. Also, just because a certain MTF limits the resolution of the image, there are very good reasons to use much higher resolution sensors to scan it. It permits noise to be removed before rescaling, producing cleaner images. The optical filtering and contrast reduction is also a major issue, since positive film materials have very high Dmax.

All true although you appear to be arguing simultaneously for limiting the capture resolution (I agree most Super-8's are not going to achieve 100lines/mm) and maximising it.

However there are diminishing returns for capturing much above 720p in terms of cost of the optics and capture device, cost of the computing power needed to process the captured data or the time to capture the data.

As it happens I don't think a DSLR is that good a capture device, yes you can lock the mirror up but if you need to do that start with a mirrorless camera. You will also get enormous frames from modern DSLRs - 24MP is not atypical - to scale that down and compress it is going to take a lot of CPU cycles. Noise reduction can be done in the time domain on lower res captures.

I'm not sure what capture device I would start experimenting with though - maybe a a webcam or a GoPro (or knockoff) might be about right.

[Planefreak]

I sort of want to stay away from using a basic video camera setup, because I'm not big on taking video apart frame by frame, to rebuild it with the correct number of frames.  I would rather deal with a couple thousand image files.  It takes my computer about an hour to render them together, create the video file, run a stabilizer on it, then compress it for a short film (~5 minutes).  Most of that time I can spend away from the computer and don't have to completely focus on it.

I've scanned a few of my films on a slide scanner.  The quality varies.  The unit captures JPG files at 3000x2000, but I was able to reduce this down to 320x240 on some of them without loss of quality.  Don't see much of point of going beyond 720p for this.

For anyone wondering, my DSLR is 18MP, but the picture size can be reduced down to 1920x1280 or 720x480 in photo mode and 1080p 30fps, 1080p 24fps, 720p 60fps, and 480p 30fps in video mode.  Not sure if I can use it in a mirror lockup in picture mode though.  When I attempt to take a picture in video mode it still closes the shutter. 

Is your purpose documentary or artistic? The usable resolution on Super8 is probably 720x480 or less, but if you want to capture the film grain effect, Kodachrome edge sharpness, etc you will need a higher resolution.

The purpose of scanning my films is to not have a bunch of film reels that are slowly rotting away falling apart.  I just want to save family movies, and retain most of the quality.  Some of the film has already been ruined, and can't be saved.

I was hoping you guys might have some ideas on making a film scanner.  I have over 12 hours of super8 film that I want to scan, but haven't come up with a truly great solution.

Take it to Costco. They have a film scanning service that will transfer it to DVD for you and save you a ton of time. Their prices are reasonable.

Looked them up and they charge $19.99 for the first 150 ft, then $0.13/ft after that.  I calculated the total cost at about $830.  I have over a mile of film.

It may be difficult to achieve frame alignment using steppers unless you have optical feedback. Even with a real projector's pull down claw, alignment can be pretty unstable. The classical way to get good alignment is to use pin registration, but there's no projector that does that to my knowledge.
If you can use optical processing, alignment may not matter. The Kinetta uses a continuous feed over a smooth drum (no gate or claw) and reconstructs frame lines digitally.

Could this be done in OpenCV?  Has anyone used OpenCV?  Seems that I could create something in there to "sense" the image, and run the control though my computer.

[helius]

I've scanned a few of my films on a slide scanner.  The quality varies.  The unit captures JPG files at 3000x2000, but I was able to reduce this down to 320x240 on some of them without loss of quality.  Don't see much of point of going beyond 720p for this.

The main reason (there are others) to scan at the maximum practical resolution, is that the image sensor in the scanner has an anisotropic screen in it. Virtually all camera sensors or scanners use rectilinear arrays of pixels, which are anisotropic. If they used hexagonal arrays, the problem wouldn't arise: when the rectilinear grid in the sensor interacts with the grain of the film, it produces larger, ugly spots. Minimizing the size of those spots means they can be more effectively removed with a noise reduction pass, such as a basic Gaussian or Median filter. It is often preferable to use LAB colorspace for this.

The purpose of scanning my films is to not have a bunch of film reels that are slowly rotting away falling apart.  I just want to save family movies, and retain most of the quality.  Some of the film has already been ruined, and can't be saved.

Preservation could also be worthwhile, although it is harder to get good information about it. To delay vinegar syndrome in triacetate films, they should be kept in cans that have a breathing hole and have chips of camphor added. Also they should be turned yearly and cleaned with an approved cleaner before storage or projection.
Film that has shrunk or warped can often be scanned using very careful handling. The Kinetta was designed to handle film very gently, so it can be used with damaged material such as old nitrate stocks. The lengths they went to are pretty impressive; it's probably beyond what you can easily do on a hobby budget.
Color reconstruction for vinegar syndrome films is sometimes possible, depending on the extent of the decomposition. The slide scanners with "Digital ROC" can do this using their fourth color channel.

Could this be done in OpenCV?  Has anyone used OpenCV?  Seems that I could create something in there to "sense" the image, and run the control though my computer.

You can overscan an area that includes the sprocket hole and use that for position detection. This is how professional bureaus usually do it. Some added image stabilization may also be used, but it has the risk of adding unpleasant "swimmy" motion if overused. OpenCV should be capable of the sprocket detection. Some modification of the projector might be required to be able to clearly visualize the hole.

[jolshefsky]

A few years ago I went and spent a lot to get the Super8 digitally scanned. The results were excellent—my goal—but it's not for everyone. I had them transferred in 2K ProRes since I only wanted to do it once and the price difference wasn't substantial, but that's way overkill. You can see grain in the 720x480 downsample, so that's plenty. The advantage of using the service was they cleaned the film, handled it carefully, scanned each frame, and provided both unaltered (ProRes) and digitally color-corrected (720p) versions. Their scanner used optical alignment, too, so even parts with sprocket hole damage didn't affect framing.

DIY, I think the modified projector is a good idea. The mechanism that keeps each frame still during projection will aid your transfer. If it were me, I'd let it run very slowly and just snap still JPEGs in sync with the gate and assemble them after the fact into film. Running the projector and trying to get digital video to line up is cheaper but will fall in- and out-of-sync.

[Richard Crowley]

I would hack an old flat-bed scanner.  Take the stepper motor that moves the scan head, and the linear image sensor.  Use the stepper motor to drive the feed sprocket, and focus the linear sensor across the film. Then you can take a high-resolution scan across the film, step the motor, rinse and repeat. With a very high resolution scan of the film (perhaps an order of magnitude greater than what is actually on the film) you can use software techniques to do image clean-up and correction, frame-jitter compensation, etc. etc.

This is the modern technique used to convert film to video. Using antique frame-by-frame projectors was made obsolete a generation ago by film scanners. Originally called "flying-spot scanners" because they used a focused CRT light source and a simple, unfocused sensor.  And they continuously pull the film through without stopping at every frame.  Much more gentle on the fragile film stock. And much easier to compensate for worn sprocket holes, etc.

There is also a technique called "wet gate" where the film is immersed in liquid of a similar refractive index to minimize the visibility of scratches and other mechanical trauma.

[Bicurico]

Hi,

The German computer magazine c't did a report on that (https://www.heise.de/ct/ausgabe/2015-7-Wie-Profis-Super-8-Filme-digitalisieren-2562344.html - you have to pay to see the full article). Apparently the best/correct way of doing that is to hack a Super8 projector:

1) using the PC or similar to have it advance slide by slide
2) sync that with a DSLR camera with remote control, attached on the projector image outlet (don't know the name)
3) replace the original lamp with a better/suitable one (can't remember which type they used, but I can look it up, if you are interested)
3) capture each slide as an individual slide
4) at the end convert it into a 24 fps movie with compression

The process is preceeded by a professional cleaning of the film. This is a really important step, as most films have collected a lot of dust and hairs/strings.

I have converted around 10 films myself half a year ago. I went the easy lane and just projected the film onto a white screen and filmed it with a mini DV camera. The results are really bad:

1) I did not clean the film, so a lot of dust, particles and fine hairs/strings collected on the corners of the image
2) The mini DV camera is filming 25 fps, but projector is rendering 24 fps - this is not nice
3) Picture is not really sharp and during play the projector actually defocusses slightly
4) Picture does not fully fill the mini DV space (4:3 rendered in 16:9)
5) The mini DV camera will capture surrounding sound (the Super8 films are mute)

Still, it is better than nothing: in my case it was old family films of me as a kid and my parents, which sadly have both died. I wanted to have it stored digitally, no matter what quality, before the equipment and/or the films would fail.

In future I will probably try again in order to get much improved results. However, there is a cost involved: a spare projector to hack, a suitable camera, new light source, controller (Arduino, Raspberry Pi, etc.), time. Probably it is cheaper to just send the films in for professional digitalisation.

Advise:

- TEST YOUR PROJECTOR/PLAYBACK DEVICE ON A SPARE FILM FIRST!!!
- Practice how to playback Super8 films on a spare film
- Purchase some cheap vintage Super8 film on eBay to do these tests
- Consider a professional conversion, but send films one by one, to make sure you don't lose them all in one go

Good luck!

Regards,
Vitor

[eliocor]

take a look at this link: http://spectrum.ieee.org/geek-life/hands-on/how-to-convert-old-film-reels-with-a-raspberry-pi
the results are rather good...
I'm thinking to do something like that.

[Planefreak]

I would hack an old flat-bed scanner.  Take the stepper motor that moves the scan head, and the linear image sensor.  Use the stepper motor to drive the feed sprocket, and focus the linear sensor across the film. Then you can take a high-resolution scan across the film, step the motor, rinse and repeat. With a very high resolution scan of the film (perhaps an order of magnitude greater than what is actually on the film) you can use software techniques to do image clean-up and correction, frame-jitter compensation, etc. etc.

This is the modern technique used to convert film to video. Using antique frame-by-frame projectors was made obsolete a generation ago by film scanners. Originally called "flying-spot scanners" because they used a focused CRT light source and a simple, unfocused sensor.  And they continuously pull the film through without stopping at every frame.  Much more gentle on the fragile film stock. And much easier to compensate for worn sprocket holes, etc.

There is also a technique called "wet gate" where the film is immersed in liquid of a similar refractive index to minimize the visibility of scratches and other mechanical trauma.

I don't think I'll get into the "wet gate" stuff.  I think I'll have enough on my hands without having to make sure I'm not going to get water everywhere.

Besides that, I am beginning to like the idea of using a scanner.  Something I see being an issue is the size of the film versus the scanning capability.  I thought most scanners only went up to 600 or 1200 dpi?  I don't think an ordinary scanner sensor would pickup the image.

[Richard Crowley]

If you have a scanner that will do 300 DPI over a width of an 8.5 inch page, then the linear sensor has a resolution of over 2500 pixels.  So if you focus the film-width onto the 2500 pixels, you get resolution well over 250px/mm which is 2-3x more resolution than in the very best film. And perhaps 10x more resolution that you can find in home-movie stock.  And scanning the entire width of the film (not just the image frame) allows you to do software correction of mechanical problems in the film like uneven exposure registration, sprocket-hole damage, film-weave, etc.

[olderbadger]

Hi,
Just joined the forum. Re: transferring S8 film to digital video (DVD). I have tried various methods and have achieved reasonable results. I have worked with film and video all my working life (now semi-retired). You need to appreciate a few things when transferring film to video/digital. 1) Frame rates 2) Aspect ratio 3) Picture stability 4) Sound synchronization 5) Capture system. OK, that said I have had success with buying a second hand (eBay) S8 projector, one with a DC motor where I could build a simple circuit to control the speed, accuracy of control is not super critical. A simple rig with a surface silvered mirror and a fine-grained rear projection screen (about A4 size). Bear in mind most S8 film would have been shot at 18 frames per second, your normal frame rate for TV(UK) is 25 fps - first dilemma. Fortunately, most S8 projectors have a 3 blade shutter so it is possible to transfer at 16 and 2/3 fps giving a slight error but livable with. If film shot at 24fps then project at 25fps (4% error) - most movies shown on TV have this error!

Key thing is to have a good projection lens / optical path and I use a Black Magic Pocket Camera, results are pretty good. With any type of modern digital 'stills' camera, you generally have a progressive shutter and you ideally need a global shutter but you can manage quite well (only fast pans can give 'Jello' effect). If projection screen texture is visible then a system can be used to vibrate or rotate this screen. Experimentation is key to decent results. If the film is damaged it may be that you can project it backwards i.e. start at the end of the film, this the uses the other edge of the perforation as a registration reference. The film can be digitally reversed in the edit before final transfer. Sound synchronisation - another can of worms, obviously not a problem with mute film. Finally, aspect ratio - Normal S8 is 4:3 ish and TV today is 16:9. My advice is centre the picture and get the height correct and then insert a mask each side to tidy the edges, this can be done in the edit. This means the finished result is showable on any 16:9 telly.

There are, of course, many other ways of transferring film but unless you want to build an expensive and complex scanner and camera system, this method works pretty well.

Good luck.

[dimkasta]

I am scanning my photo films using an Epson v600 with the relevant 35mm and 60mm adapters and the result is superb.

It will be quite some work to scan so many frames though. But there are some adapters and software that may help you scan many frames at a time.

[james_s]

The old fashioned film to TV transfer machines used a flying spot scanner. This consisted of a CRT with a special low persistance phosphor which the film was passed over and a photomultiplier tube to pick up the light passing through the film. The scanning CRT was driven with the same sync signals used to generate the video output and the output from the PM tube became the luminance signal. It's possible something like this would work with one of those tiny camcorder viewfinder CRTs, it's not the right phosphor but it could be fun to try.