The result of a little experiment I conducted a while back: a simple algorithm that attempts to identify scenes in a video file that will combine to produce a mosaic of a target image.
It works surprisingly well for such a poor algorithm: lots of quality data + lots of time + low quality algorithm = acceptable results.
Observe that the algorithm attempts to match detail within each frame in an attempt to increase the effective resolution of the resulting mosaic.
Interestingly, the tendency of the algorithm to favour a good overall luminance match gives rise to its greatest weakness: a tendency to repeatedly use the same frames.
Daisy Garden, my latest Android application, in action. A poor quality video unfortunately - shot from my webcam. It doesn’t include sharing flowers yet, but it will.
