Abstract: A restoration method for video sequences based on distinguishing damaged objects was proposed. hhc restoration of the first damage is based on total variation algorithm, while the second damage is based on texture synthesis from samples with spatial-temporal correlation. For the former, the eight neighborhoods of target point in scratches and small speckle are repositioned. For the later, the search of match blocks is limited in the 21、21 range with repairing block-centered in front and back frames. The experimental results show that the method can improve effectively efficiency and effectiveness for video sequences' restoration. It is applicable for roboticized restoration of old film material.

Key words: Video restoration, Scratch, Total variation, Neighborhood, hexture synthesis