Super-resolution from a low- and partial high-resolution image pair

Abstract

The classical super-resolution (SR) setting starts with a set of low-resolution (LR) images related by subpixel shifts and tries to reconstruct a single high-resolution (HR) image. In some cases, partial observations about the HR image are also available. Trying to complete the missing HR data without any reference to LR ones is an inpainting (or completion) problem. In this paper, we consider the problem of recovering a single HR image from a pair consisting of a complete LR and incomplete HR image pair. This setting arises in particular when one wants to fuse image data captured at two different resolutions. We propose an efficient algorithm that allows to take advantage of both image data by first learning nonlocal interactions from an interpolated version of the LR image using patches. Those interactions are then used by a convex energy function whose minimization yields a superresolved complete image.

Image Zoom Completion

“…Indeed, the former is based on a nonlocal graph where each missing pixel is connected to a set of pixels that lie in the HR region, while the latter adds to the previous connections a set of connections from each pixel to its nearest neighbors in a local window. In this respect, the present construction generalizes that of [2]. We also note that the optimization algorithm used in the present paper is different from that of [2].…”

“…A preliminary version of this paper has been published in [2]. We bring to the attention of the reader the fact that the nonlocal regularization proposed in [2] is different from the present one.…”

“…We bring to the attention of the reader the fact that the nonlocal regularization proposed in [2] is different from the present one. Indeed, the former is based on a nonlocal graph where each missing pixel is connected to a set of pixels that lie in the HR region, while the latter adds to the previous connections a set of connections from each pixel to its nearest neighbors in a local window.…”

“…Here, the problem is to recover a high-resolution (HR) image from a pair consisting of a complete low-resolution (LR) image and an HR but incomplete one. To the best of our knowledge, this problem has been introduced for the first time by the authors of the present paper in [2].…”

Super-Resolved Enhancement of a Single Image and Its Application in Cardiac MRI

“…Once the mapping function is learned, it is applied to a single testing image to achieve SR. Despite numerous learning or example based methods (e.g., [7,8,9,10]) claiming success for single-image SR, these methods hinge on the availability of training data of LR and HR image pairs; therefore, we define them as pseudo-single-image SR.…”