Unsupervised SAR image segmentation based on kernel TMFs with belief propagation

Abstract

The triplet Markov field (TMF) model has achieved promising results in synthetic aperture radar (SAR) image segmentation. Focusing on the simple likelihood modelling of an SAR image and the effective optimisation of the TMF model, an unsupervised SAR image segmentation algorithm based on kernel TMF with belief propagation is proposed in this study. Instead of complex speckle noise statistical models in a spatial domain, a piecewise constant likelihood model in the kernel mapped space is used in the TMF model by mapping the SAR image data into a higher dimension space via a kernel function. The max-product belief propagation algorithm is used to realise effective optimisation of the proposed kernel TMF model. Experiments on both simulated and real SAR images demonstrate the effectiveness of the proposed algorithm.

Keywords

• synthetic aperture radar
• speckle
• maximum likelihood estimation
• image segmentation
• markov processes
• radar imaging
• simulated sar images
• real sar images
• triplet markov field model
• synthetic aperture radar image segmentation
• simple likelihood modelling
• effective optimisation
• unsupervised sar image segmentation algorithm
• complex speckle noise statistical models
• piecewise constant likelihood model
• kernel mapped space
• sar image data
• higher dimension space
• kernel function
• max-product belief propagation algorithm
• kernel tmf model