Automatic Registration of High Resolution SAR and Optical Satellite Imagery in Urban Areas

Kurzfassung

The work presented in this thesis addresses the problem of registration of high resolution SAR and optical satellite imagery in dense urban areas. With the launch of high resolution remote sensing satellites in different modalities like TerraSAR-X, WorldView-1 and IKONOS-2, the contribution of remote sensing for various applications has received a tremendous boost. Specifically, the combined analysis of high resolution SAR and optical imagery is of immense importance in monitoring and assessing catastrophes and natural disaster. Although, latest satellites provide georeferenced and orthorectified data products, still registration errors exist within images acquired from different sources. These need to be taken care or through quick automated techniques before the deployment of these data sources for remote sensing applications. Different registration methods like the ones based on detecting geometrical features like interest points and line detectors to intensity based strategies have been rightfully evaluated. Considering the meticulous task of extracting and matching conjugate features in SAR and optical imagery (especially metric resolution imagery) a general feature based image registration technique for various scenarios might be difficult to develop and implement and therefore region based approach has been selected in this thesis. In general, for high resolution satellite images acquired over urban areas, common city features like wide roads, rivers, big stadiums, play grounds, parks can be expected to appear in considerable sizes and be represented by relatively homogeneous intensity values and therefore the explored strategies in this thesis are based on region detection which in particular show a robust performance for high resolution SAR images. The scheme here is to detect 'on ground' regions, which are not expected to be affected by very different SAR and optical sensor geometries. The following step selects the extracted regions and prepares them for matching with methods based on cross-correlation and mutual information. Selected datasets for the testing and evaluation of the region based registration methods are pre- and post-disaster (2008 earthquake in Sichuan China) IKONOS-2 images along with a TerraSAR-X image acquired within hours after the disastrous earthquake. The evaluated automated methods in this thesis have successfully detected shift parameters within already orthorectified SAR and optical imagery acquired over dense urban areas.