Abstract: To solve image super-resolution problems, in a learning-based framework, clustering method and the Gaussian Process Regression are employed.The k-means algorithm is used to make data clustered.And in each cluster, the relationship between the low-resolution images and the high-resolution images is modelled by the Gaussian Process Regression through the learning in the training dataset.With the learned model, the high-resolution image can be obtained by given a input low-resolution image.The statistical relationship between the low-resolution images and the high-resolution images is effectively utilized and the results demonstrate the proposed algorithm can make improvement in super resolution reconstruction.