Classification of a polarimetric SAR image using fractal concepts

Assesses the use of fractal dimensions as discriminators in SAR images for different terrain types. A preliminary study is performed using an L-band polarimetric SAR image. Two procedures for obtaining fractal dimensions from SAR data at three polarizations HH, VV, and HV are presented. The results show that the fractal dimensions which depend upon the spatial correlation function and the distribution function of the backscatter intensity are less dependent on the polarization. Classification performance using fractal dimensions is compared to that obtained by backscatter intensities for broad classes of terrain. Significant improvements have been obtained, especially in the case of an homogeneous image texture, despite the diversity of mean backscatter intensities, in an area. The results are promising and suggest further studies to assess the robustness of fractal dimensions as discriminators in various SAR images of different terrain types