Another option would be to...

Another option would be to...

Posted Dec 14, 2007 18:47 UTC (Fri) by jd (guest, #26381)
Parent article: The Grumpy Editor's video journey, part 1

...regard the analog medium as simply a collection of stills. Scanning in a time series of still images is much much easier in Linux. You even have the added benefit that if you rescan the same still using red, green and blue, there are packages for stitching such images together into a single high dynamic range image, which may be closer to what was originally on the analog medium.

Next up, if there is a sound track, you record that. It's kept separate from the images, at this point you only care that both are time sequenced.

You then thread the time-sequenced stills and the sound into a single movie, where the inputs are scaled to an equal length in time, hopefully but not necessarily linearly. That's the hard part. When you record the sound, the slice of sound for a given frame must be shown at the same time as that frame. The painfully bad sync on some YouTube videos, where the individuals involved aren't doing anything nearly as sophisticated as what I'm outlining, shows that this is not an easy thing to do.

You now (hopefully) have something that accurately reproduces the movies, is likely superior to something Linux' video input can do (because you produced multiple samples of the images and that can't be done in software alone in most cases), requires much simpler drivers (the graphics aren't sampled in hard real-time and aren't obtained from high-speed video capturing hardware), and will probably cause far less damage to the hair follicles.

If the series doesn't already cover such approaches, then maybe it might be worth adding something on non-traditional video capture methods - in this case, using the fact that video is really no different than animation via flicking through pages in a book.

---

(Log in to post comments)

Sorry, what are you talking about?  Taking the raw analog tape out of the video and laying it
down on a flat-bed scanner (which makes no sense), or... what?

Not quite, though I didn't spell out what would be done. My fault there, sorry. The idea is to
step a single frame at a time in the movie projector, and capture that frame as though it was
a single still. It doesn't matter whether it's direct or a reflected image. You'd use an image
captre device, like a monochrome CCD, but since we're not playing the frames at regular speed,
the capture device can operate much slower and capture much more detail. The capture device
also wants to be as far from the projector as produces more useful detail.
<p>
(This eliminates blur, allows you to get all of the detail of the frame, etc.)
<p>
If it takes 1 second to capture a still at such high resolution, you simply forward to the
next frame about once a second, and each frame becomes photographable. (The BBC used similar
techniques to produce production stills from camera footage.)
<p>
You do this once with red only light from the camera, then again with only green light, and
finally with blue light. Hence the use of a monochrome CCD. You get a higher resolution
capture, but also by blending the images, you end up with HDR stills. Not sure if there is an
animated version of OpenEXR, but there is a version of MPEG specifically for HDR. You then use
that to convert the stills into a movie.
<p>
I'm probably still skipping details, but it's a variant on how they "clean" analog movies
(which is done frame-by-frame) and then use those stills to rebuild a full movie, originally
also on tape (so they had to burn a single frame onto a single celluloid window, with very
high accuracy).

but the rest of the conversation is how to get from analog recorded vidow, not from film.

doing a freeze frame on videotape is not practical.