Stochastic and Deterministic Methods in Motion Picture Restoration

In this paper we apply stochastic methods (Gibbs sampling and simulated annealing) and deterministic (mean field) approximations to the problem of detecting and interpolating `dirt and sparkle' or `blotches' in archived motion pictures. 1 Introduction The problem addressed here is that of restoring `scratches' or `blotches' which are a major defect in archived motion pictures. These can be formed by abrasion removing some of the emulsion, causing a bright defect, or dirt becoming attached to the film, resulting in a dark defect. The common feature of these defects is that they can be characterised as a temporal discontinuity. In an image sequence, the scratches and blotches are obvious as areas for which there is no match in the previous or subsequent frames, taking motion between the frames into account. The heuristic for the detection of scratches and blotches is thus -- a scratch is a connected area of a frame which is unpredictable from either the previous or following frames. The..