A groupwise super-resolution approach: Application to brain MRI

Image super-resolution techniques provide a route to studying fine scale anatomical detail using one or more lower resolution acquisitions. A crucial issue in such algorithms is the form of image regularization used to constrain the image structure at points where there are insufficient data values. In this paper we examine the specific problem of reconstructing a high resolution isotropic image when presented with a set of low-resolution anisotropic images. In particular here, we propose to extend recently proposed patch-based methods for super resolution to this problem. More specifically, we develop regularization term which is designed to take advantage of information redundancy in the set of images. We include an experimental evaluation using the MR Brainweb database and a comparison which shows significantly improved reconstruction details when compared to conventional interpolation based methods.