Superpixel segmentation of PolSAR images based on improved local iterative clustering

Abstract

ABSTRACT In order to solve the problem that the simple linear iterative clustering (SLIC) cannot segment the noisy image well, the local iterative clustering (LIC) is improved and successfully applied to the superpixel segmentation of polarimetric synthetic aperture radar (PolSAR) image. However, this algorithm still has two problems: the computation of edge graph is too large and the content of superpixel cannot guarantee the homogeneity. In this article, the LIC algorithm is improved to segment PolSAR image better and faster. For the first problem, the algorithm introduces the integral graph into edge graph calculation to remove the redundant operation; for the second problem, the algorithm adds hypothesis testing to test the homogeneity of superpixel content, and re-segment the superpixel that does not meet the homogeneity hypothesis. The experiment is carried out with the measured data of AIRSAR system and GaoFen-3 system, and the experimental result shows that the algorithm can double the calculation speed of edge graph and significantly improves the superpixel segmentation effect.