SAR Image Segmentation Based on Mixture Context and Wavelet Hidden-Class-Label Markov Random Field

Because of the property that synthetic aperture radar (SAR) image includes plenty of multiplicative speckle noise, a new segmentation algorithm is proposed based on the mixture context and the wavelet hidden-class-label Markov random field (MRF). In our paper, a wavelet mixture heavy-tailed model is constructed, then the hidden-class-label MRF is extended to the wavelet domain to suppress the effect of speckle noise. Since the multi-scale segmentation with overlapping window is presented here, this segmentation method is utilized at the finest scale of stationary wavelet transform (SWT) domain, and the classical segmentation method is also utilized at the coarse scales of discrete wavelet transform (DWT) domain. Finally, a mixture context model is proposed to estimate inter-scale parameters and the optimal segmentation result is derived from a new maximum a posteriori (MAP) classification. The experimental results show that our method achieves accurate SAR image segmentation result and preserves detailed boundary effectively.