PERFORMANCE EVALUATION OF A COMBINED MULTI-RESOLUTIONTEXTURAL CLASSIFICATION OF A SINGLE ERS-1 IMAGE

摘要

Against the multispectral images, working only on onernsingle monobande Radar image, for cost and imagernavailability reasons, is a hard task due to nature of thesernimages. In this work, we present an attempt for thernclassification of an ERS-1 Radar image for the need ofrnthematic cartography. The drawback of monobandernRadar image is surmounted by the integration of thernspatial dimension of the image i.e. the texturalrninformation. This precious information is obtained byrnusing the wavelet transformation. This latter provides usrnthe possibility to extract the maximum of texturalrninformation by introducing the changing in imagernresolution in the textural analysis scheme. A redundantrnwavelet transformation provides us a ‘tower’rnrepresentation instead of the pyramidal one, thus, wernpreserve pixels to pixel correspondence between therndifferent towers level. A textural information fusionrnstep is then applied in order to deal with the well knownrn‘curse of dimensionality’ problem.rnA supervised classification process, which combinernboth radiometric and textural information’s, in a multiresolutionrnscheme is applied. The results obtained overrna single ERS-1 image which cover a relatively complexrnarea are satisfactory. This permits us to confirm that thernpresented classification scheme is a good approach forrnthe purpose of cartography with only a single Radarrnimage.