Improving blotch detection in old films by a preprocessing step based on outlier statistical test

This work deals with the detection of blotches which are specific degradations commonly affecting old movies. The novelty of our approach consists in integrating a preprocessing step into any conventional blotch detector operating in the spatial domain. This preprocessing step aims at outputting candidate regions that may be degraded. Its originality relies on the fact that it is based on a statistical outlier detection test coupled with a compensation motion method. Experimental results carried out on real old movies, show that this preprocessing highly improves the detection performances of the conventional Simplified Rank Ordered Differences (SROD) detector and the detector based on the Auto-Regressive (AR) model.