A Novel Decision-level Fusion Segmentation Approach of SAR and Optical Images

A novel decision-level fusion segmentation approach of two Synthetic aperture radar (SAR) images and one optical image is presented. First, the energy function for optimal segmentation of two SAR images is deduced and fusion segmentation approach based on the Markov random field (MRF) model for SAR images is proposed. Then, an optical image is employed to further improve the segmentation performance and accelerate the segmentation process of two SAR images. By using of optical image classification, the image pixels are classified into three classes: certain target pixels, certain background pixels and uncertain pixels, respectively. The certain pixels' labels are maintained as their ultimate segmentation labels. Only the uncertain pixels need to be segmented by SAR fusion segmentation method based on the MRF model. Finally, a fast annealing strategy is proposed to accelerate the fusion segmentation process of uncertain pixels. Results of computer simulation validate the effectiveness of the proposed method.