Level Sets Based on AoTF Fitting Energy for Riverbank Line Extraction From SAR Images

This article proposed a level set method (LSM) based on a novel data fitting energy for extracting riverbank lines from synthetic aperture radar (SAR) images. First, we devise a function by computing the absolute-value of the tangent function, which we call AoTF, and employ it to establish the data fitting energy. Second, the refined area fitting centers (AFCs) are developed in terms of two types of AFCs, namely average and median AFCs, which obtain better accuracy and stability. Third, a difference of Gaussian (DoG) based edge-indicator is used to supplant the Dirac-function in the model’s gradient descent flow (GDF), which encourages the evolving level sets to locate at the target edges. In addition, some regularizing constraint terms are incorporated into the objective function. Riverbank line extraction experiments on actual SAR images demonstrate that the proposed LSM outperforms some state-of-the-art methods in extraction performance and is robust to the level set initialization.