Download the complete thesis in pdf format (in French)
ABSTRACT
Most motion pictures produced before the fifties have a very short life span. They are very damaged and their condition continues to deteriorate. It is necessary, and urgent, to restore and preserve them. Besides, if these films were restored, they could be used to feed the rapidly increasing audio-visual market.
Classical restoration methods can correct some defects, but not all. Computer based restoration methods are paid more attention, but most of them treat the frames manually, one by one: results are good, but still very expensive. In order to restore a greater number of motion pictures, new faster techniques must be developed. One way of doing this is to increase the degree of automatism in the restoration process.
This PhD thesis is among the first to propose automatic restoration algorithms for old motion pictures. We give below a list of the defects that we treat, as well as a short description of the corresponding restoration algorithms.
Flicker
This defect corresponds to an abnormal variation of the luminance in the scene. In order to eliminate it, we limit the variations of the frames histograms along time.
Jittering
We propose a simple method for evaluating the global translation between consecutive frames. Next, we filter the sequence of translations in order to separate the vibrations component from the natural movement of the scene.
Vertical scratches
We detect vertical scratches frame by frame using morphological operators. We then spatially interpolate the missing grey levels.
Local random defects
All defects that cover a relatively small region of each frame and that most of the time do not appear in the same position between consecutive frames belong to this category.
We detect them by using spatio-temporal connectivity criteria, implemented through openings and closings by reconstruction.
However, when there is fast motion in the scene, these criteria may find problems. In order to solve them, we have developed a new motion compensation algorithm. It is based on a segmentation of the reference image, and uses kriging in order to interpolate and filter the displacement vector field.
Interpolation
Once the detection phase is completed, lost grey levels have to be recovered. We have developed spatial and temporal interpolation methods to restore the missing pixels.
A complete restoration system
Finally, all these algorithms have been embedded in a restoration system that we have called SARSA, which stands for Système Automatique de Restauration de Séquences Animées (Automatic Restoration System for Animated Sequences). We have used it to restore severely damaged sequences with success.
Bibliographical reference
E. Decencière Ferrandière. Restauration automatique de films anciens. PhD thesis, Ecole Nationale Supérieure des Mines de Paris, December 1997.