Self-content super-resolution for ultra-HD up-sampling

• DOI: 10.1145/2414688.2414695
• Publication Date: 2012-12-19T09:12:22Z
• Authors: Mehmet Türkan; Dominique Thoreau; Philippe Guillotel
• Keywords: Content (measure theory)

We describe a self-content single image super-resolution algorithm based on multi-scale neighbor embeddings of patches. make use the recurrence property similar patches across different scales an image. Inspired by manifold learning approaches, we first characterize local geometry given low-resolution patch reconstructing it from taken down-scaled versions input then hallucinate high-resolution relying geometric similarities low- and spaces. enforce compatibility through overlapping, preserve structures with sparsity-based averaging. further global consistency back-projection. Noting that this method uses as little self-information contained in its implementation is well suited to GPU processors thanks highly parallelization, our experimental results demonstrate or even better performance respect state-of-the-art superresolution methods.