A wavelet-based texture feature set applied to classification of multifrequency polarimetric SAR images

Abstract

Texture is an essential key to the classification of land cover in SAR images. A wavelet-based texture feature set is derived. It consists of the energy of subimages obtained by the overcomplete wavelet decomposition of local areas in SAR images, where the downsampling between wavelet levels is omitted. The feature set has been successfully applied to multifrequency polarimetric images of the Flevoland site, an agricultural area in The Netherlands. The methods of polarization selection and feature reduction are also discussed.