Automatic coral island segmentation based on region-based multi-phase level set method: A case study on Pattle Island, South China Sea

Remote sensing provides an effective way to observe and monitor coral islands with shallow coral reefs worldwide, to characterize inter-reef structural differences, and to map intra-reef zonation. Nowadays, although types of high and moderate resolution remote sensing data become various, the existing feature extraction methods are mainly confined in data-driven classification field. To maintain spatial-continuity of features, this paper applies a model-driven, multi-phase level set method (MPLSM) for specific circle-structured coral island segmentation based on CBERS-02B CCD multi-spectral image. The fundament of multi-phase level set method is to establish local cluster criterions to control level set evolution. Experiment shows that: (1) MPLSM segmentation avoids topology error and contour omission problems compared to classic Chan-Vese piece constant model; (2) MPLSM is robust to contour initialization so as to improve level set automation; (3) MPLSM cannot handle with the diffusive boundary and areas seriously influenced by mixed-pixel effect, but can ensure level set convergence.