Restoration of digitized video sequences: An efficient drop-out detection and removal framework

Millions of hours of valuable audiovisual content is endangered or already destroyed. Today, restoration is mainly done manually, which is time-consuming, costly and thus simply infeasible for large amounts of data. For this reason, automation of restoration efforts is of major importance to win the race against time. In this paper, a framework for efficient drop-out detection and restoration is presented. This artifact class is one of the most frequently occurring in video archives. The proposed detection algorithm is a two-pass approach, where frames of the potentially deteriorated video sequences are classified into valid and suspect based on global color statistics of the images. Suspect pictures are further submitted to local, quad-tree-based analysis for refined evaluations. This yields a subset of identified damaged pictures with accurately localized defects. Detected defective frames are restored using a motion compensation-based approach. Experiments on a data set based on video sequences of the ¿PrestoSpace¿ project show very promising results.