Target Detection and Texture Segmentation in Polarimetric SAR Images Using a Wavelet Frame

Abstract:
A technique is presented to detect structures and segment texturally homogeneous regions in SAR images using features extracted by a wavelet frame decomposition. The technique is also extended to the polarimetric case. To the purpose polarimetric synthesis is performed in the wavelet domain. A formalism is introduced, called the Wavelet Polarimetric Signature, that encapsulates in graphical form the global dependency of the textural features on the polarization configuration. Controlled experiments based on simulated SAR images and related to detection of weak but polarimetric diverse targets in clutter are reported. Next a test case using real SAR imagery and in connection with surfactant slick detection in coastal waters is described. Results of the experiments confirm that the combination of textural features and polarimetric diversity yield improved performance in various detection and segmentation scenarios.