Extracting characteristic segments in high-resolution panchromatic imagery as basic information for object-driven image analysis

For the monitoring of urban areas, high-resolution aerial image data are still primarily used to detect individual man-made structures. Image segmentation techniques extract and provide basic information such as regions of similar features or typical object edges for further model-driven analysis. Therefore, the quality of the final object recognition depends strongly on this information. In this paper, the potential of a combined edge- and region-based segmentation technique for the purpose of detecting characteristic man-made objects is tested and compared to standard techniques using a new quality measure. The presented approach proves its value, especially for objects with low-contrast boundaries.