Superpixel Segmentation for PolSAR Images Based on Hotelling–Lawley Trace Distance

Abstract:
As a preprocessing step, segmentation accuracy and runtime are integral criteria of the superpixel segmentation for PolSAR images. The Hotelling–Lawley trace (HLT) test statistic satisfies the three well-defined criteria of the measure and complies with the complex Wishart distribution, which is advantageous to measure the pairwise similarity. However, in order to maintain the symmetry, the amount of computation is doubled for calculating HLT distance. To solve this problem, this paper proposes to combine the HLT distance with geodesic distance (GD) to reduce the time consumption. The GD derived from the real symmetric Kennaugh matrix is capable of measuring the shortest distance between pixels with higher efficiency. Experimental results show that the superpixels generated using the proposed method exhibit higher segmentation accuracy and lesser runtime, compared to five other state-of-the-art methods.