Underwater object highlight segmentation in SAS image using Rayleigh mixture model

Image segmentation, which usually employs a statistical model, is an essential step in synthetic aperture sonar (SAS) image processing. This work addresses the Rayleigh mixture model (RMM), representing SAS underwater amplitude image. High resolution SAS image of detected artificial object is segmented using RMM and Markov random field (MRF) model. We present a quick unsupervised iterative method to segment the object (highlight). In each iteration, RMM parameter is estimated by EM algorithm, and used by graph-cut based MRF image segmentation. The algorithm converges, and gives the final segmentation. Experiment on real SAS data shows that the RMM is capable of describing complex object echo distribution, thus improve the segmentation quality of SAS image.