Level sets with one dot fuzzy initialization for marine oil spill segmentation

Xu, Min and Yu, Yunhua and Chen, Fang and Jiang, Xiangyuan and Ren, Peng and Yang, Erfu; (2017) Level sets with one dot fuzzy initialization for marine oil spill segmentation. In: OCEANS 2017 - Aberdeen. IEEE, GBR. ISBN 9781509052790 (https://doi.org/10.1109/OCEANSE.2017.8084923)

[thumbnail of Xu-etal-IEEE-Oceans-2017-Level-sets-with-one-dot-fuzzy-initialization-for-marine-oil-spill-segmentation]
Preview
Text. Filename: Xu_etal_IEEE_Oceans_2017_Level_sets_with_one_dot_fuzzy_initialization_for_marine_oil_spill_segmentation.pdf
Accepted Author Manuscript

Download (1MB)| Preview

Abstract

Image segmentation techniques are widely used for identifying marine oil spill regions in SAR images, with one representative technique being level set evolution. The accuracy of level set evolution highly relies on a proper initial level set function which roughly captures the marine oil spill layouts in a SAR image. However, marine oil spill regions are always complicatedly shaped, and it is time-consuming and impractical to manually devise precise initial level set functions for various marine oil spill shapes. In order to address this problem and achieve efficient and robust segmentation, we develop a one dot initialization scheme for level set evolution. Our method just requires an arbitrary pixel within one marine oil spill region as initialization. It exploits fuzzy connectedness between pixels such that consistent initial level set functions can be established via arbitrary initial dots within the marine oil spill region. Experimental results validates the robustness of our strategy.