Robust Feature Based Multisensor Remote Sensing Image Registration Algorithm

Abstract

The crucial problem of multisensor remote sensing image registration is how to establish the reliable correspondences between the features extracted from two images. The feature similarity based methods fail when similar local regions exist, and the spatial relationship methods fail to match small portion of pair wise correspondences out of the total number of features. In this paper, we proposed shape context as feature similarity and overlap error as spatial relationship to construct an objective function for the feature matching problem, and deterministic annealing (DA) is used to solve the optimization problem. Feature matching experiments with real remote sensing images demonstrate the superiority of our algorithm over the 5 classic feature matching algorithms, and registration results outperform the popular image registration algorithms.