Automated Non-Gaussian Clustering of Polarimetric SAR

Abstract:
This paper presents an automatic image segmentation method for Polarimetric SAR data. It utilises the full polarimetric information and incorporates texture by modelling with a non-Gaussian distribution. The modelling is based upon the well known product model, with a Γ distributed texture parameter, leading to the K-Wishart model for the covariance matrix. The automatic clustering is achieved through a modified Expectation Maximisation algorithm, with an additional Goodness-of-fit test allowing splitting and merging of clusters. The resulting image segmentation depicts the statistically significant clusters within the image. Real world examples are shown to demonstrate the technique.