Ratio- and distribution-metric-based active contours for SAR image segmentation

Abstract

A novel active contour model driven by local and global intensity fitting energy is proposed in this paper. The model overcomes the defect of some well-known active contour models which tend to fall into local minimums in SAR (synthetic aperture radar) image segmentation. Firstly, a new ratio distance, which measures the relativity between two speckle-image patches, is defined by using the probability density functions of the regions inside and outside the contours. Then, a new image energy functional is computed with the ratio distance, and the region intensity fitting functions of CV (Chan and Vese) active contour model are replaced by the new image energy function. Finally, the proposed model is constructed based on the combination of the improved CV model and RSF (region-scalable fitting) model by linear weights. The model makes the contour evolution more efficient with the LIF (local intensity fitting) force and the GIF (global intensity fitting) force, which are respectively provided by the RSF model and the improved CV model. SAR image segmentation results validate the effectiveness of the proposed model.