Unsupervised change detection frameworks for very high spatial resolution images
Fabio Pacifici, Chris Padwick, Giovanni Marchisio · 2010
Two different unsupervised change detection techniques are here investigated. The first method is based on pulse-coupled neural networks, which show invariance to object scale, shift or rotation. The second method, based on the normalized cross-correlation, is suited to work in an “on-line” processing as more images are made available, for example in case of natural events such as an earthquake or tsunami. The performances of the algorithms have been evaluated on pairs of QuickBird, WorldView-1 and WorldView-2 images taken over Atlanta (U. S. A.), Washington D. C. (U. S. A.), and Conception (Chile).