Parametric and Nonparametric Methods for SAR Patch Scene Categorization

This paper presents synthetic aperture radar (SAR) image categorization based on feature descriptors within the discrete wavelet transform (DWT) domain using nonparametric and parametric features. The first and second moments, Kolmogorov Sinai entropy and coding gain, are used for the nonparametric features within an oriented dual tree complex wavelet transform (2D ODT$\mathbb{C}$WT). A Gauss-Markov random field (GMRF), triplet Markov random field (TMRF), and autobinomial model (ABM) are used for feature extraction using a parametric approach within an image domain. A single parameter of GMRF, TMRF, or ABM is used for characterizing an entire patch; therefore, higher model orders (MOs) are used. A database with 2000 images representing 20 different classes with 100 images per class is used for estimating classification efficiency. A supervised learning stage is implemented within a support vector machine (SVM) using 10% and 20% of the test images per class. The experimental results showed that the nonparametric features achieved better results when compared to the parametric features.