Supervised High-Resolution SAR Image Segmentation Algorithm Based on Level Set

A new supervised level set segmentation method based on statistics model for high-resolution synthetic aperture radar (SAR) images is proposed. The target and background scattering statistics characteristic of the high-resolution SAR images is modeled by Fisher and Gamma probability density function separately, and an energy functional with respect to level set adapted for SAR image is defined. Partial differential equations (PDE) of curve evolution are obtained by minimizing the energy functional. Meanwhile, the parameters of the Fisher and Gamma distribution are estimated by training data selected in advance. The segmentation of the SAR images is implemented by the solution of the PDE. The performance of the method is verified by real SAR images. Results show that the method can get faster segmentation speed and more rounded target segmentation for targets with strong reflectors of high-resolution SAR images if only the training data are selected suitably.