Super-resolution mapping using the two-point histogram and multi-source imagery

A new method for super-resolution classification from remotely sensed imagery is presented. The method allows prediction of a super-resolution (sub-pixel) land cover map from a coarse spatial resolution (original pixel) land cover proportions image and an intermediate spatial resolution panchromatic (Pan) image. The method is based on spatial simulated annealing and combines two objectives: (i) to match a prior sub-pixel two-point histogram obtained from some training image and (ii) to match the predictions made of an intermediate spatial resolution panchromatic image from the super-resolution classification via a forward model to an observed panchromatic image. The method is demonstrated on simulated remotely sensed imagery. The main advantage of the Pan image is to fix locally the outcome of the two-point histogram objective function.