Multiscale identification of buildings in compressed large aerial scenes

The growing amount of images to be processed for scientific or intelligence purposes makes the use of compression algorithms quite frequent. We show that it is possible to localize objects in images compressed with quincunx wavelets. The quincunx multiscale edges of both the object and the scene are first computed, and then matched using a modified Hausdorff distance. The localization begins at coarse resolution and continues through the finer ones. Computing time is short on a workstation, and implementation on a parallel computer is possible, so that real-time computation is plausible Robustness of the method is also proven