Movie restoration by topological asymptotic analysis

You can download my talk (6th workshop on Mathematics and Image Analysis, September 2006, Paris):
Image restoration and classification by topological asymptotic analysis
if you want a brief overview of the mathematical study and to see why we have a complexity in n.log(n) (where n is the size of the image/movie).
Note that this is an old presentation, you can contact me for a newer one (or have a look at my habilitation thesis webpage).

You can download here the corresponding (avi) movies. The numerical experiments have been performed on a laptop (2.0 GHz, 1 GB RAM), with Matlab 7.0, without optimization.
In C, on a faster laptop, the computation time has been divided by more than 1000.
The matlab, C and C-GPU codes are available under request.

• Size of the movie: 180x72 pixels, 30 frames => 388 000 points, computation time: 5 minutes.
  • Original movie (150 kB)
  • Noised movie (300 kB)
  • Restored movie (150 kB)
  • Detected contours (300 kB)
  • All together (1.4 MB)

• Size of the movie: 288x176 pixels, 52 frames => 2.6 million points, computation time: 1 hour.
  • Original movie (120 kB)
  • Noised movie (500 kB)
  • Restored movie (120 kB)
  • Detected contours (600 kB)
  • All together (1.7 MB)

• Size of the movie: 320x144 pixels, 100 frames => 4.5 million points, computation time: 2 hours.
  • Original movie (200 kB)
  • Noised movie (1.7 MB)
  • Restored movie (260 kB)
  • Detected contours (2.5 MB)
  • All together (4.8 MB)

• Size of the movie: 512x288 pixels, 110 frames => more than 16 million points, computation time: 12 hours.
  • Original movie (1.3 MB)
  • Noised movie (5.9 MB)
  • Restored movie (1.1 MB)
  • Detected contours (7.6 MB)
  • All together (5.6 MB)

• Synthetic movie, with a missing frame, and no noise. Computation time: a few seconds.
  • Original movie (6 kB)
  • Noised movie (6 kB)
  • Restored movie (6 kB)
  • Detected contours (10 kB)
  • All together (220 kB)