A Comparison of Patch-Based Models in Video Denoising

Several state-of-the-art patch-based methods for video denoising rely on grouping similar patches and jointly denoising them. Different models for the groups of patches have been proposed. In general more complex models achieve better results at the expense of a higher running time. But the modeling of the groups of patches is not the only difference between the approaches proposed in the literature. Other differences can be the type of patches, the search strategies used for determining the groups of similar patches and the weights used in the aggregation. This makes it difficult to determine the actual impact of the patch model on the results. In this work we compare two of the models that have produced better results in equal conditions: those assuming sparsity on a fixed transform (like BM3D), against methods that seek to adapt the transform to the group of patches. In addition we propose a third simple model which can be interpreted as a non-local version of the classical DCT denoising and add it to the comparison. We compare the three models with 3D large patches and use the optical flow to guide the search for similar patches, but not to shape the patches. Either one of the three approaches achieves state-of-the-art results, which comes as a consequence of using a large 3D patch size. As expected, the adaptive transform attains better results, but the margin reduces significantly for higher noise levels.