Assessment of two Gibbs random field based feature extraction methods for SAR images using a Cramer-Rao bound

Abstract:
This paper is an assessment of two feature extraction methods based on Gibbs random fields models using a Cramer-Rao Lower Bound (CRLB) and a Fisher Information Matrix (FIM). The first evaluated method is the model-based despeckling and information extraction algorithm using a Gauss-Markov random field (GMRF) as prior. The second one is the maximum a posteriori auto-binomial method (MAP-ABM), which uses an auto-binomial model as prior. The assessment has been carried out using simulated SAR data. In here, data with an increasing number of looks have been used in order to study 1) how the estimated parameters approach the real ones, 2) how their variances get closer to the CRLB, 3) which are the best model parameters, and 4) how to make a comparison between both model parameters. The experimental results show the superiority of GMRF parameter estimation. Both GMRF and MAP-ABM provide the most robust texture parameters when the number of look is between 2 and 4. MAP-ABM parameter estimation presents a dependence on the size of the estimation windows compared with GMRF.