Medical image super-resolution via minimum error regression model selection using random forest

Super-resolution is designed to construct a high-resolution version of a low-resolution for more information. Super-resolution can help doctors to get a more accurate diagnosis. In this paper, we propose a novel super-resolution method utilizing minimum error regression selection. In the training step, we partition the patches into multiple clusters through jointly learning multiple regression models. Then we train a random forest model based on the patches of multiple clusters. During the reconstruction step, we use trained random forest model to select the most suitable regression model for the reconstruction of each low-resolution patch. Several medical images are applied to test the proposed method. We compare both the objective parameters and the visual effect to other state-of-the-art example-based methods. Experiment results show that the proposed method has better performance.