Squeezing the soul out of digital video

The image above is from the original Teenage Mutant Ninja Turtles cartoon title sequence. Pretty iconic, right? It is the result of a new video technique I came up with. For more examples and a thorough explanation, read on:

I was taken by a Strange Mood and created a small combination of shell and python scripts that:

1) Creates a still image from every frame of a given input video, then 2) Compares each of these images against each other, round-robin style, in order to 3) Find the two images (and therefore, the two “shots”) which are the LEAST like each other in the source video.

Essentially it take a video input, and finds the two frames that are least like each other. My theory is that all of this will Tell Us Something. I don’t really know what. This is something like digital mysticism, trying to find the soul of a string of bits and surface it.

The current method is sub-optimal in several ways, for one it takes a long time to run on a laptop. Remember: We’re comparing every second of video to every other second of video, and that adds up. Running the script against a full 22-minute episode of a TV should would require 970921 comparisons, so I’ll set that up to run tonight and maybe it’ll be done by morning? This sounds like a job for EC2.

Some more examples:

A-Ha – Take On Me

Citizen Kane – Original Trailer


FFMPEG Script To Save Off An Image Every Second, Minute, Etc.

ffmpeg -i input.flv -vf fps=1 out%d.png

Very useful snippet on the ffmpeg site. You’ll see.