Master thesis at the University of Stuttgart
In this thesis, suitable image features to solve correspondence problems were learned via Covolutional Neural Networks. These features are embedded in the Coarse-to-Fine PatchMatch method to estimate the optical flow between two images.