Video shot detection and characterisation in semi-automatic digital video restoration

In film archives there is a lot of material not exploitable due to its bad condition. In semi-automatic approaches to digital video restoration the operator browses the video and bounds, around damaged frames, video sections on which to apply appropriate sequences of spatio-temporal filters. Smoothness of change in scene properties around the damaged frames is essential for proper operation of filters. We suggest using video indexing techniques to automatically identify violations of interframe spatio-temporal continuity and to support the operator visual pre-processing. We report and discuss the results of a preliminary test performed on indexing a sample sequence of an old degraded film. We obtain high defect recall and precision indexes spending in pre-processing just a few percent of the average time required for digital restoration