Along with the rapid development of modern medical imaging technologies and computational resource, high dimensional medical images play more and more important roles in clinical applications. However, there still present many modeling and computational challenges for restoring and processing medical images with high accuracy and efficiency in practice. In this talk, I will first introduce sparsity promoting and redundancy exploring approaches for some classes of image restoration problems, such as multi-modality image and dynamic images. Then I will present our recent deep learning approaches for MRI restoration and dual modality processing.