Image Classification by means of CEM Algorithm based on a GARCH-2D data model

Abstract

The aim of synthetic aperture radar (SAR) classification is to assign each pixel to a class according to a feature of the illuminated area. In this work, a classification method suitable for SAR images is presented through the maximum a posteriori (MAP) criteria by means of the expectation-maximization (EM) algorithm based on a mixture of GARCH-2D processes data model. This model assumes that the data probability density function (pdf) is a combination of a finite number of pdf's of GARCH-2D processes, that represent the pixel classes and whose parameters are estimated iteratively by means of the EM algorithm. Once the parameter estimation is performed, the a-posteriori probability of each pixel belonging to each class is computed and the classification is performed through the MAP criteria. Based on this model, the expressions for estimation and classification procedures are derived. Finally, the method performance is verified through a numeric example for a particular case and a comparison is performed between this approach and a variant of the classification algorithm based on a Gaussian mixture model for the data pdf.