Multiscale quadtree model fusion with super-resolution for blocky artefact removal

Abstract

Digital elevation model (DEM) integration by optimally fusing multiple measurements having different characteristics can play an important role in estimation and prediction of environmental changes and natural disasters. Multiscale or multiresolution modelling used for this purpose has attracted attention in terms of its rich modelling capability as well as computational efficiency. In particular, a multiscale Kalman smoother (MKS) with tree structure graph can provide a powerful, efficient algorithm based on the Markov property. However, due to the stair-like correlation from the tree structure, unrealistic artefacts can be generated in the fusion result. This is especially prominent on geographic surfaces or ocean surfaces that have naturally smooth characteristics. In this article, a super-resolution (SR) algorithm is applied to remove blocky artefacts. Using the algorithm, new measurements are generated in the area where blocky artefacts can arise. The measurement is then fused with vector-valued MKS. The results show that this novel image fusion scheme provides a smaller root mean square error (RMSE) as well as better visual inspection.