A Comparative Study of Two Multiresolution Methods for the Mapping of a Large Coastal Zone Area from a Synthetic Aperture Radar Image

Abstract

The problem which worries us in this study is the mapping of an Atlantic coastal zone from a synthetic aperture radar (SAR) image of very great size. It is known that reducing the size of very great size images by the multiresolution techniques allows to resolve problems of slowness when such images are processed with a weak capacity computer. However, reducing the size of an image causes a loss of information. In this study, the technique of reducing images by the Burt's method and the technique of wavelets by the Haar's method are compared, using as an example, an unsupervised classification of a SAR ERS-1 image on the Cameroonian Atlantic coast. The loss of information of the image is estimated as it is reduced, for each of these methods. This loss of information shows itself by an alteration of the ground's occupation area of the various classes resulted from the unsupervised classification. An extrapolation of the results is then made to allow a valuation of the ground's occupation surfaces on the original image which was not able to be processed because of its very great size. Within the framework of this study, the method of the Haar's wavelets offers a better stability of the results compared with the Burt's method and it allows to have a good valuation of the real ground's surfaces of the various classes of the image.