A Fast Algorithm for Single Image Super-Resolution Reconstruction via Revised Statistical Prediction Model

A fast algorithm of single image super-resolution reconstruction using a revised statistical prediction model based on dictionary learning has been proposed in the paper. The statistical prediction model uses MMSE estimator to promote performance cascading several levels of the basic algorithm. Although this approach has advantage than the existing approaches that based on a pair of low and high resolution dictionary. These approaches have computational complexity for training dictionary. The revised model focuses on handling the single image based on recognition model. The single image is divided into two parts: the main information part and the background part. The recognition model is utilized to find the region of main information of a single image. The main information is only designed for learning parameters to reduce the complexity. The background image is processed by Tikhonov Regularization super resolution reconstruction, which reduce the running time in spite of having bad edges and textures. The proposed algorithm provides lower computational complexity and running time than the statistical prediction model, when completing the dictionary training.