Improving spatial and spectral resolution of satellite images

In this paper we present a new approach of image fusion to di erent spatial and spectral resolutions. Our approach uses the supervised trainning to nd the appropriate space representation. The aim is to generalize the di erent nonlinear color space transformations. Thus, this allows to generalize perceptual methods. To compare our approach with existing methods we have used the assessment techniques of the image fusion quality. A generic protocol that treats the consistency and synthesis property has been retained. This requires the application of the extrapolation hypothesis and the global quality index. The di erent methods have been tested on extracts acquired by the IKONOS and the QuickBird satellites. The result show a slight improvement of perceptual methods in comparison with others. Also, we have compared the HSV transformation with our approach. It emerges of this study that the result obtained by our approach are slightly better.