Resource information
Super-resolution mapping (SRM) is a process that provides land-cover spatial distribution with a spatial resolution finer than the size of a remotely sensed image pixel. Usually, the fraction images generated from soft classifiers are used as constraints in SRM, making the accurate estimation of fraction images an important task in SRM. Supervised fuzzy c-means (sFCM), which belongs to the fuzzy-set technology commonly applied for unmixing mixed pixels, is capable of providing accurate estimates of fraction images used for SRM. It can also generate more than one output of fraction images with a different degree of fuzziness, providing more flexible choices for the fraction images for SRM. However, unmixing errors of sFCM are carried over to SRM, which uses fraction images as a constraint. In this letter, a novel sFCM-based SRM (sFCM_SRM) model that incorporates the pixel-unmixing criterion of sFCM in its objective function is proposed. Without using fraction images as a constraint, sFCM_SRM can be directly applied to remotely sensed images and can eliminate the influence of pixel-unmixing errors on the resultant super-resolution (SR) map to some extent. Both synthetic multispectral and IKONOS images were applied, and the proposed sFCM_SRM model was evaluated by comparison with the SRM model in which sFCM was first used to generate fractions and then pixel-swapping algorithm (PSA) was applied to yield the final SR map. The sFCM_SRM model generates a more visually and quantitatively accurate SR map.
