A Fast and Accurate Cluster Center Initialization Algorithm for PolSAR Superpixel Segmentation
Binbin Han, Ping Han, Zheng Cheng · 2019
Locate iterative clustering (LIC) gets good performance in superpixel segmentation for PolSAR images. However, there are also two problems in its cluster center initialization. One is that there is a lot of computational redundancy in calculating edge map, which slows down the segmentation speed. The other is that the calculated edge is much wider than the real one, which makes the fine-tuning of center seeds in 3 × 3 window not suitable. To solve these problems, a fast and accurate cluster center initialization algorithm for PolSAR superpixel segmentation is proposed in this paper. For the first problem, integral graph is introduced to eliminate the redundancy. For the second problem, a more reasonable window size is selected to conduct fine-tuning. Experiments with measured AIRSAR data sets demonstrate the speed of the proposed edge map calculation method increases 28 times and the center seed is farther away from the edge than before. Furthermore, LIC gets better superpixel segmentation results with the new initialization algorithm.