Skip to Main Content
In this paper, we present a novel example selective and order independent method for learning-based image super-resolution. We first present a method that selectively utilizes training samples according to the content of the input image. Experimental results show that by selecting the training samples appropriately, it is possible to dramatically reduce the computational costs without degrading image quality. We then present a new order independent technique that is shown to perform better than traditional order dependent techniques in learning image super-resolution and can also be applied to image editing such as region filling and object removal from images.