1950: The year nitrate-based film

was completely discontinued,

due to its habit of exploding

DIY Film Digitizer

A modified projector
can convert old
movie reels

y grandfather Leo was

M a self-taught electrical
engineer and IEEE mem-
ber who designed control systems
for tire factories. He was also an
avid photographer, and his eight
children—and later, his grand-
children—were among his
­favorite subjects, right up to his
death in 1974, when I was 5.
Fast‑forward to 2013: ­During
a move, my uncle ­uncovered
a trove of more than 130 reels
of Leo’s 8-mm and 16-mm
home movies, some dating back
to 1939. While commercial conver-
sion services exist, converting so many
reels would have been pretty expensive,
so my cousin and I set out to preserve them
•
digitally ourselves. First, a quick primer on film pro-
jection for digital natives. In a projector, a motor pulls film
through a vertical “gate.” Each film frame is held still and
flat in the gate while a lamp illuminates it from within the
projector’s housing. Lenses on the other side of the gate
focus the image so that it appears sharp on whatever sur-
face the film is projected. Between the gate and the lamp, a
rotating shutter wheel blocks the light while the next frame
is sliding into position. (Without this shutter, the film would
Joe Herman

Frames to Files: A film projector can be

modified to capture old movies by reversing •
be one big blur.) For our initial movie conversion attempt,
the path that light travels through it. we tried to record digital video d ­ irectly from a projector.

SPECTRUM.IEEE.ORG | North American | jan 2017 |  17

01.Resources.NA.indd 17 12/13/16 6:06 PM

 RESOURCES_HANDS ON

We mounted a DSLR camera to a series of Movie MODS Excited by this success, I started capturing
lenses focused on the gate. We started the films, but it soon became unbearably tedious.
projector and camera, and everything seemed I had to run the system more slowly than I ex-
to be going great. The camera recorded video pected (under a frame per second) to ensure
with excellent detail and c­ olor. However, it also consistent behavior from the motor and ac-
highlighted the dirt, cracks, and scratches on commodate delays caused by the Pi’s slow
many of the reels. Even more ­problematic, file system. I also had no way of easily viewing
the speed of our ­projector—as with most— images and adjusting settings midcapture,
couldn’t be precisely controlled. This meant so I found myself having to recapture entire
that even under the best conditions, mis- reels. Most worrying, after a few reels it was
matches between the frame rates of the pro- clear that the Pi camera’s low dynamic range
jector and the camera resulted in rolling dark relative to that of the film was wiping out a lot
areas, flickering, and other artifacts. These, of detail in high-contrast scenes.
along with the dirt and scratches, could not A year (and lots of programming) later,
be easily remedied digitally. though, I’ve managed to address each of these
Going back to the drawing board, we learned issues. To deal with the most critical—the high
that a frame-by-frame transfer—capturing a dynamic range of the film—I employ the same
still image of each film frame and then convert- approach as smartphones do in similar cir-
ing the collection of images into a video—gives cumstances. I shoot multiple images per frame
far superior results. Still images are easily ame- with different exposures (two or three is usually
nable to digital postprocessing. Such transfers enough) and combine them into a single image
are usually performed on specialized equip- using functions available in the OpenCV library.
ment for high fees. Fortunately, I remembered This requires significant processing, so
there was a spare Super 8 projector in my base- instead of saving images on the Pi, I stream
ment and an unused Raspberry Pi in my closet, them via Ethernet to a PC running Linux
and that a Pi camera was just US $30. Could desktop. This also eliminates the bottleneck
we do a frame-by-frame transfer with these? of the Pi’s file system. I further increased cap-
I removed the projector’s motor and shut- ture speed by swapping out my original mo-
ter wheel. I connected a small 60-rpm 12-volt tor for a more reliable NEMA 17 stepper
motor to the projector’s drive shaft using a sec- motor. Using my fastest settings, I can now
tion of thin fuel hose. Some basic interfacing capture at more than two frames per second.
electronics let me control the motor’s speed I built a GUI control panel for the Linux ma-
and direction with the Pi. Then I attached a chine using PyQt, which lets me monitor cap-
small magnet to the drive shaft so that a reed tures on the fly and adjust settings.
switch would close each time the film ad- Postcapture, I can quickly recombine
vanced one frame. This switch was connected a reel of images into a movie using one
to the Pi so that it would trigger the Pi camera. By replacing a projector’s lamp and ­FFmpeg command. When advanced resto-
The Pi camera’s strength is that it’s very easy original motor with a Raspberry Pi ration is needed, I use AviSynth, a free video-­
to control using Python software libraries. This and a stepper motor [top] frames can processing tool for which others have written
be recorded using a camera [second
is a huge asset for a project requiring precise from top] located in the housing and scripts tailored specifically for film restoration.
timing and color and exposure adjustments. illuminated from the outside via the I have now built this setup three times:
But the Pi camera’s fixed-focus lens isn’t de- projector’s lenses [second from bottom]. on 8-mm, Super 8, and 16-mm projectors.
A reed switch detects when the next frame
signed for close-up photography, and replac- of film is ready to be captured [bottom]. On the 16 mm I was able to embed the Pi,
ing it with a different lens produces s­ evere camera, and motor into the lamp housing
color vignetting. Luckily, by carefully unscrew- to create a portable capture rig. I’m still de-
ing the lens nearly all the way out of its housing, tor’s housing, so I placed my LED light source veloping the software, but so far I’ve cap-
I was able to bring the Pi camera’s focal dis- “backward,” shining down the projector’s tured about 7 hours of footage—nearly half
tance down to about 20 mm. By adding an ad- original lens toward the film. Placing a dif- a ­million frames—and the results are beauti-
ditional magnifier lens, I could get a nice sharp fuser (a bit of translucent white plastic from ful. It has been a long project, but being able
Joe Herman (4)

image of an 8-mm frame in the gate. an old ceiling fixture) as close to the film as to revisit—and share—such precious ­family
Positioning the camera close enough to possible helped to hide scratches and age- history has made it more than worthwhile.
the gate required mounting it in the projec- related cracks in the film’s coating. —Joe Herman

18  | jan 2017 | North American | SPECTRUM.IEEE.ORG

01.Resources.NA.indd 18 12/13/16 6:06 PM