Sea-Land Segmentation Algorithm for SAR Images Based on Superpixel Merging

摘要

Aimed at improving the accuracy of sea-land segmentation for synthetic aperture radar (SAR) images, a novel sea-land segmentation approach based on superpixel merging is proposed. First, pixel dissimilarity measure is defined on the basis of simple linear iterative cluster (SLIC) algorithm and image presegmentation steps are given to generate superpixels to prepare primitive subregions for the following merging process. Second, three rules, including vicinity rule, edge rule, and similarity rule are proposed to guide the superpixel merging and then a coarse-fine merging strategy is presented to complete sea-land segmentation. Experimental results show that the proposed method has higher segmentation accuracy for SAR images compared with other algorithms.