A structural low rank regularization method for single image super-resolution

摘要: Example-learning-based algorithms such as those based on sparse coding or neighbor embedding have been popular for single image super-resolution in recent years. However, affected by several critical factors the training data and example representation, their reconstructions are usually plagued kinds of artifacts. The removing these artifacts is one major tasks methods. Unlike most existing methods that employ more complicated methods, this paper we would like to recover a clear reconstruction fusing "dirty" coarse which outputs simple with small set. One underlying key observation although corrupted different artifacts, they refer same high-resolution image. This global structure information captured an structure-based low rank regularization method. advantage our method it can remove not only noises but also gross Except sparsity randomness large no other knowledge about them required. Experimental results show proposed dramatically improve achieve competitive results.