Extended multi-level logistic model and SAR image segmentation

An extended multi-level logistic (EMLL) model and its parameter estimation method is proposed in this paper. The multi-level logistic (MLL) model is an extendedly used markov random field model. When using MLL model for label image, there is an underlying supposition that all boundaries have the same characters. This supposition is not always correct. The proposed EMLL model solves the drawback of MLL model by associating different parameters with different boundaries. EMLL model has much more parameters than MLL model. A linear parameter estimation method for EMLL model is proposed and the experimental results for synthetic label images show that it performs well. We use EMLL model and its parameter estimation method in MAP segmentation of SAR images and the segmentation results show that EMLL model performs better than MLL model.