Markov Networks for Super-resolution Markov Networks for Super-resolution

We address the super-resolution problem: how to estimate missing high spatial frequency components of a static image. From a training set of fulland lowresolution images, we build a database of patches of corrsponding highand low-frequency image information. Given a new low-resolution image to enhance, we select from the training data a set of 10 candidate high-frequency patches for each patch of the low-resolution image. We use compatibility relationships between neighboring candidates in Bayesian belief propagation to select the most probable candidate high-frequency interpretation at each image patch. The resulting estimates of the high-frequency image are good. The algorithm maintains sharp edges, and makes visually plausible guesses in regions of texture. Published in Proceedings of 34th Annual Conference on Information Sciences and Systems (CISS 2000), Dept. Electrical Engineering, Princeton University, Princeton, NJ 08544-5263, March, 2000 This work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonpro t educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of Mitsubishi Electric Information Technology Center America; an acknowledgment of the authors and individual contributions to the work; and all applicable portions of the copyright notice. Copying, reproduction, or republishing for any other purpose shall require a license with payment of fee to Mitsubishi Electric Information Technology Center America. All rights reserved. Copyright c Mitsubishi Electric Information Technology Center America, 2000 201 Broadway, Cambridge, Massachusetts 02139 1. First printing, TR2000-08, March, 2000 Egon Pasztor's present address: MIT Media Lab 20 Ames St. Cambridge, MA 02139 2000 Conference on Information Sciences and Systems, Princeton University, March 15-17, 2000 Markov Networks for Super-Resolution W. T. Freeman and E. C. Pasztor MERL (Mitsubishi Electric Research Lab) 201 Broadway, Cambridge, MA 02139 e-mail: [email protected], [email protected] Abstract | We address the super-resolution problem: how to estimate missing high spatial frequency components of a static image. From a training set of fulland lowresolution images, we build a database of patches of corrsponding highand low-frequency image information. Given a new low-resolution image to enhance, we select from the training data a set of 10 candidate high-frequency patches for each patch of the low-resolution image. We use compatibility relationships between neighboring candidates in| We address the super-resolution problem: how to estimate missing high spatial frequency components of a static image. From a training set of fulland lowresolution images, we build a database of patches of corrsponding highand low-frequency image information. Given a new low-resolution image to enhance, we select from the training data a set of 10 candidate high-frequency patches for each patch of the low-resolution image. We use compatibility relationships between neighboring candidates in Bayesian belief propagation to select the most probable candidate high-frequency interpretation at each image patch. The resulting estimates of the highfrequency image are good. The algorithm maintains sharp edges, and makes visually plausible guesses in