Couleur et texture pour la représentation et la classification d'images satellite multi-résolutions

Land use mapping and characterization are very important for local and national institutions. These institutions are nowadays searching for specifie and specialized tools that can distinguish betweendifferent land covers. This research work proposes to use different methods for satellite image processing. Allowing a strong and reliable land cover classification. The conceptual and experimental design has been developed as it follows. First, an optimal description of ail images is done. Then, COIOL and texture attributs are defined and computed. Finally, sorne algorithm classifications are realized. The optimal description of ail images is made by (i) determination of the hybrid colour space to obtain a good discrimination of this classes while correlation between space components is mininized. (ii) merging a high spatial resolution panchromatic image with a low spatial resolution multispectral image in order to obtain a high spatial and spectral resolutions image. Attributes are then extracted to characterize land cover classes using colour and texture information through different approaches (statistics, geometry, frequency, fractal, multifractal). At last, different classification techniques are applied (SVM, MMG, K-means, ISODATA) in order to separate forest areas from agriculturc parcels. Our work originality is based on the construction of a hybrid colour space derived from the image intensity, saturation and hue omponents using a multiobjective approach that integrates the correlation and discriminating power. This same space ls used in the merging images process in order to aeneralize the perceptual methods.