532 of very high spatial resolution satellite image Comparison segmentations.pdf

Comparison of very high spatial resolution satellite image segmentations A. Carleer a , O. Debeir b , E. Wolff* a a Institut de Gestion de lEnvironnement et dAménagement du Territoire, Université Libre de Bruxelles, cp 130/02, Avenue Franklin Roosevelt, 50, 1050 Brussels, Belgium; b Information and Decision Systems, Université Libre de Bruxelles, cp 165/57, Avenue Franklin Roosevelt, 50, 1050 Brussels, Belgium ABSTRACT Since 1999, very high spatial resolution data represent the surface of the earth with more details. However, information extraction by computer-assisted classification techniques proves to be very complex owing to the internal variability increase in land-cover units and to the weakness of spectral resolution 1, 2, 3 . The increase in variability decreases the statistical separability of land-cover classes in the spectral space 4 . Per pixel multispectral classification techniques are then insufficient for an extraction of complex categories and spectrally heterogeneous land-cover, like urban areas 5 . Per region classification was proposed as an alternative approach 6, 7 . The first step of this approach is the segmentation. A large variety of segmentation algorithms were developed these last 20 years 8 and a comparison of their implementation on very high spatial resolution images is necessary. For this study, four algorithms from the two main groups of segmentation algorithms (boundary-based and region-based algorithms) were selected. In order to compare the algorithms, an evaluation of each algorithm was carried out with empirical discrepancy evaluation methods. This evaluation is carried out with a visual segmentation of IKONOS panchromatic images. Keywords: segmentation, very high spatial resolution satellite images, evaluation methods. 1. INTRODUCTION The first commercial very high resolution satellite (IKONOS) became accessible in autumn 1999 and the QuickBird satellite in October 2001. These new sources of ver