Accuracy improvement and evaluation measures for registration of multisensor remote sensing imagery

Abstract

Intensity based image registration methods are widely used in fine geometric registration of multisensor images. Accordingly, for images that are compared through translation of image templates, position where similarity measure is maximized is assumed to indicate best registration. Image registration quality is of crucial importance especially for studies that have high geometric accuracy requirements; e.g. image fusion, change detection, multichannel segmentation, and Digital Terrain Model (DTM) generation. Accuracy of image registration is conventionally evaluated by means of error measures (e.g. RMSE) obtained through comparison of coordinates of control points from the target and the reference / ground truth. However, especially for multisensor images with low spatial resolution component, difficulty in precisely positioning control points inhibits both sub pixel accuracy and evaluation of the registration. In this study, three widespread measures in intensity-based image registration namely, Normalized Cross Correlation (NCC), Mutual Information (MI), and Phase Correlation (PC) are tested for registering images acquired from EO-1 Hyperion and IKONOS sensors. We propose the use of ‘global similarity’ and ‘inverse consistency’ measures for evaluating the performance of these intensity based automated registration methods.