Jäger, Marc; Hellwich, Olaf (Computer Vision and Remote Sensing, Berlin University of Technology, Germany)
Images contain information at a range of different scales, or levels of detail: scenes contain objects of various sizes and individual objects are single semantic units at one scale and decompose into object parts at finer scales. Scale spaces enable the extraction of information at arbitrary scales, and are consequently useful in the automatic interpretation of images. The present paper defines scale spaces for polarimetric SAR data, and describes mechanisms for information extraction. The applications discussed include the scale invariant detection of distinctive regions, which frequently correspond to characteristic parts of objects, and a segmentation algorithm to estimate the exact extent of the structures that prompted region detection.