High performance motion analysis for video restoration

We explore the use of motion estimation methods to restore noisy motion pictures coming from degraded film in the framework of high-performance computing environments. Specifically, we investigate how the use of block-matching methods can outperform gradient-based methods (which are more sensitive to little movements) also in the case of blotch detection in degraded film. Also, we show how a previously proposed method characterized by the use of a set of features together with multiresolution can be at the same time robust to noise and effective in motion compensated filtering, mainly of concern in films that contain both large amounts of noise and fast/large motion objects in their frames. The advantages and disadvantages of the methods are discussed and the model performances are assessed and compared with examples.