Superpixel Driven Unsupervised Deep Image Super-Resolution

Most of the existing deep learning-based image super-resolution methods require a large number datasets or ground truth. However, these are not suitable for restoration real with different domains. Recently, Deep Image Prior (DIP) based on single-image explores prior and uses network structure as implicit to recover images, but it ignores explicit information actual itself. The addition can effectively alleviate ill-posed problem in model. Therefore, this paper, we propose an unsupervised (SR) method that segmentation driven. Intuitively, clear has clearer boundary. It will drive neural networks obtain higher performance SR when forcing restored have In order make energy flow into DIP better, use fully convolutional networks-based superpixel method, back propagation inject gradient generated by entropy lower optimization parameters. Experiments show our boundary better than Set5, Set14 BSD100.