Sea ice segmentation of SAR imagery using multi-temporal and multi-scale feature extraction

A method for sea-ice segmentation of SAR imagery is proposed in this paper. It is based on the multi-location descriptor extraction in salient image points, which indicate image locations with temporal change, high intensity contrast, regional homogeneity and shape saliency. A novel spatiotemporal operator is used to extract the feature points, which advantageously integrates temporal change with spatial saliency. The final segmentation by scale-adaptive region growing provides borders of sea ice regions. Preliminary experimental results confirmed the reliability of the proposed segmentation method and have shown its high potential for SAR image analysis.