When I watch old films, and when I say “old films,” I refer to movies dating back to the 1920s and earlier, and I experience feelings of nostalgia and glamour. Those feelings are transmitted to me by means of the movements of the people in those films, which are not so smooth as nowadays; the costumes and poses of the actors, which were much more theatrical in those days; and also by the degradations of the reel itself, with spots due to dirt attached to the film, and abrasion of the emulsion or line scratches, usually generated by rubbings against rotating mechanical pieces. These artifacts are very undesirable for automatic processing by computers.
This paper provides a new technique for avoiding line scratches in old films. The authors provide an algorithm divided in two stages: a detection step, based on Kalman filtering and Hough Transform, and a removal step, based on a Bayesian model. This removal step has two sub-steps: a reconstruction of low frequencies by a simple polynomial interpolation, and a refinement by MAP reconstruction in which similar film grain is added to give a smoother aspect, in order to reconstruct high frequencies damaged by line scratch.
In this paper, the authors claim that line scratch removal is a fast process: ten seconds per video image in the detection step and two seconds per line scratch in the removal and restoration step on a standard 350 Mhz Pentium II. It is worth noting that each frame is scanned at 2200 x 1640 pixels at 12 bits/pixel, which makes every frame grow up to approximately 43.7 megabytes. However, in spite of the size of each frame, it is too pretentious to claim that the method is fast, because it is rather far away from real time processing.
Another disadvantage is the lack of a robust set of video sequences: just two film sequences have been used with up to 140 frames. And finally, although visually the whole process produces very good results, there is no mathematical measure for the quality of these results. In any case, this paper is a nice approach to the very difficult task of historical film restoration.