Complex-valued Markov Random Field Based Feature Extraction for InSAR Images

In this paper, complex-valued Markov random field (CMRF) parameters, namely the interaction strength and variance, which have been previously used for noise reduction in interferograms, are proposed for feature extraction from interferometric SAR (InSAR) images. A comparative performance evaluation has been carried out for feature extraction from InSAR and single-look complex (SLC) SAR images. A patch-based classification is performed for a small database of 3 forest classes. Also, a single image is tiled into small patches and unsupervised clustering is performed. The results are compared to that of another MRF-based complex-valued feature vector which consists of complex-mean and covariances.