A Segmentation Method of SAR Images Based on Non-parametric Density Estimate and Markovian Contexture

Aiming at giving a precise estimation of the statistic distribution of high-resolution Synthetic Aperture Radar (SAR) images, a segmentation method of SAR images using technique of non-parametric density estimate with kernel method and Markovian contexture is proposed in this paper, after studying the traditional models based on parametric technique. First, a non-parametric density estimate method based on kernel function is adopted to estimate the statistic distribution of the SAR images, and then, the SAR images is segmented with Markovian contexture by maximizing a MAP estimator, taking the former estimation as its likelihood term. And the results of the new proposed method and methods based on parametric statistical models are compared by software simulation. It shows that non-parametric density estimate technique based on kernel function can provide better results by just depending on real data, when there is no available analytical distribution function. Experiments on real SAR images also show that the non-parametric method can model the complex scenes of high-resolution SAR images such as urban areas well and get better results of segmentation.