Spatial continuity and self-similarity in super-resolution mapping: self-similar pixel swapping
Yuan-Fong Su · Remote Sensing Letters · 2016
Self-similarity of fractal geometry refers to that a part of an object is similar to the whole. This scale-invariant feature has a certain role to play in super-resolution mapping which is a mapping technique across scale aiming at enhancing spatial resolution of remote-sensing imagery. Unlike other super-resolution mapping methods depending solely on spatial continuity, a self-similar pixel swapping (SSPS) method combining spatial continuity and self-similarity of fractal geometry into original pixel swapping (PS) algorithm is presented here. A self-similar weight function defined from the composition information at pixel scale within a predetermined window is added to the calculation of attractiveness in the standard PS method. The self-similar weight function guides the subpixels within a pixel to arrange spatially similar to the appearance of the composition information at pixel scale. Evaluating with synthetic images and satellite image, the performance of the SSPS is particularly obvious in reproducing objects with sharp corners, linear features and adjacent small objects.