VHR satellite image segmentation based on topological unsupervised learning

Nistor Grozavu, Nicoleta Rogovschi, Guénaël Cabanes, Andres Troya-Galvis, Pierre Gançarski

Abstract: High spatial resolution satellite imagery has become an important source of information for geospatial applications. Automatic segmentation of high-resolution satellite imagery is useful for obtaining more timely and accurate information. In this paper we introduce a new approach for automatic image segmentation into different regions (corresponding to various features of texture, intensity, and color) based on topological un-supervised learning. Three types of methods were studied in this work: matrix factorization, self-organizing maps and probabilistic models. The approaches were applied on a real Very High Resolution (VHR) image of the French city of Strasbourg. The obtained segmentation results were validated using internal and external clustering validation indexes.