Stable underwater image segmentation in high quality via MRF model

The underwater image segmentation in high quality is one of the important parts in ocean investigations. Markov random field (MRF) based framework has been considered as one powerful statistical estimation approach for the spatial continuity on the basis of the Bayesian theory. However, in the underwater imaging system, poor visibility in the sea and the variations in the illumination, viewpoints, etc., have limited the application of the standard MRF and cause the sensitivity to the speckle noises in underwater image segmentation. Hereby, in this paper, we try to explore one kind of the underwater image segmentation scheme combining the MRF model and Hard C-means(HCM) clustering technique with the regional labeling and the local characteristics concerned, achieving great performances in both robustness and effectiveness.