Unsupervised Segmentation of SAR Images Based on Multi-feature Fusion and Hidden Markov Tree
Guan Xiao-ping · Remote Sensing Information · 2006
We present a new unsupervised segmentation algorithm for SAR images based on hidden Markov tree(HMT) using multi-feature fusion.The multiscale gray value data and standard variance data of the image are first modeled with HMT respectively.Then according to the two segmentations obtained the structure information is extracted from them.The posterior marginal probabilities of the gray feature HMT are smoothed using the structure information.For single look SAR images,where Rayleigh distributions are considered appropriate for each class,we also give four parameter estimation algorithms for HMT.The experimental results show that the unsupervised approach can give a more smooth segmentation than HMT,and meanwhile preserving the structure information well.