Object-based Normalized-Cuts for High-Resolution Polsar Image Segmentation

Xiaofang Xu,Bin Zou,Lamei Zhang

DOI: https://doi.org/10.1109/BIGSARDATA.2019.8858494

2019-01-01

Abstract:As more and more high-resolution PolSAR images being available, effective methods need to be studied for the processing of high-resolution PolSAR images. Since the complexity and speckle noise in high-resolution PolSAR image, traditional pixel-based image segmentation methods can result in heavy salt-and-pepper noise and the segments won't be in accord with the edge in PolSAR images. Object-based image analysis (OBIA) methods have been proven to be effective for high-resolution images. In this paper, a novel object-based image segmentation method for high-resolution PolSAR image is proposed, combining the superpixel algorithm and Normalized-Cuts segmentation (NCuts). The proposed method first generates original objects using simple local iterative clustering (SLIC) which is a new superpixel algorithm; and then Ncuts segmentation is implemented on the original objects to obtain refined object segments and reduce the computational complexity. In addition, both spatial information and polarimetric information will be introduced in the proposed method. Experiments using EMISAR image show that the proposed method can better preserve the edges and homogeneous regions on the image.