A Novel Example-Based Super-Resolution Approach Based on Patch Classification and the KPCA Prior Model

In this paper, we propose a novel example-based super-resolution method to hallucinate high-resolution images from low-resolution images. As example-based super-resolution is a kind of learning process, how to learn effectively from training samples is essential to the quality of the reconstructed images. In our algorithm, a classification process is firstly employed to construct a well-organized patch database. Then, the KPCA prior model is used for each class to infer the high-resolution output. Since the training samples or patches are divided into numerous classes, the variations among the patches in each class or cluster are therefore greatly reduced. In addition, KPCA can capture the high-order statistics in those training samples, which makes the learning process even more powerful. Experiments show that the proposed algorithm can provide a high quality for image super- resolution reconstruction.