On the Possibility of Intensity Based Registration for Metric Resolution SAR and Optical Imagery

Kurzfassung

Multimodal image registration is a key to many remote sensing tasks like fusion, change detection, GIS overlay operations, 3D visualization etc. With advancements in research, intensity based similarity metrics namely mutual information (MI) and cluster reward algorithm (CRA) have been utilized for intricate multimodal registration problem. The computation of these metrics involves estimating the joint histogram directly from image intensity values, which might have been generated from different sensor geometries and/or modalities (e.g. SAR and optical). Modern day satellites like TerraSAR-X and IKONOS provide high resolution images generating enormous data volume along with very different image radiometric properties (especially in urban areas) not observed ever before. Thus, performance evaluation of intensity based registration techniques for metric resolution imagery becomes an interesting case study. In this paper, we analyze the performance of similarity metrics namely, mutual information and cluster reward algorithm for metric resolution images acquired over both plain and urban/semi-urban areas. Techniques for handling the generated enormous data volume and influence of really different sensor geometries over images especially acquired over urban areas have also been proposed and rightfully analyzed. Our findings from three carefully selected datasets indicate that the intensity based techniques can still be utilized for high resolution imagery but certain adaptations (like compression and segmentation) become useful for meaningful registration results.