Segmenting multisensor aerial images in class–scale space

We introduce a class–scale space for automated segmentation of aerial images from diverse sensors. We describe the performance of a segmentation algorithm operating in this space. This algorithm is carried out by three processors in succession: multiscale feature extractor, multiclass pattern classifier, and class–scale logic. The multiscale feature extractor extracts features at three levels of precision. The multiclass pattern classifier is an array of neural classifiers, one classifier for each class–scale pair. Each classifier is optimized by genetic feature selectors and genetic weight initiators. The array of neural classifiers produces an array of segmented images. Class–scale logic operates on these images to produce a final-segmented image that combines coarsely detected regions with finely detected curves and points. We describe applications of these techniques to segmenting and matching SAR, IR, and visible-light images.