V.A. Franz, V.V. Voronin, V.I. Marchuk, M.M. Pismeskova

The task of video repairing is related to the problem of image inpainting. The only difference is the necessity to maintain temporal continuity in addition to the spatial continuity. The first work in this area have used information from neighboring frames for recovering procedure. This approach is justified in removing the defects of the film. Many types of defects appear only on one frame, and absent from his neighbors. The virtue of this method is its simplicity. But it is not suitable for delete an object that is presented in several successive frames. In this paper we propose a new method of reconstruction of video sequences based on the modeling scene and track objects in the scene. Proposed approach allow to remove objects or restore missing or tainted regions present in a video sequence by utilizing spatial and temporal information from neighboring scenes. The algorithm iteratively performs following operations: achieve frame; update the scene model; update positions of moving objects (this step use the condensation algorithm); replace parts of the frame occupied by the objects marked for remove with use of a background model. We demonstrate the performance of a new approach via several examples, showing the effectiveness of our algorithm and compared with state-of-the-art video inpainting methods.