Semantic segmentation of polarimetric SAR image via the uniform homogeneity slice

PolSAR image segmentation has long been an important problem in the PolSAR remote sensing community. Many segmentation algorithms describe images in terms of a hierarchy of regions. However, it is often over-segmentation at finer levels and under-segmentation at coarser levels, which makes it difficult to adopt the hierarchies for later use. In this paper, we propose an effective measure to extract a single segmentation from the hierarchical structures of PolSAR images. We call our method the uniform homogeneity slice, which select a set of regions that balance the homogeneity from the hierarchy. First, we construct the Binary partition tree (BPT) which is a multi-scale image representation to obtain a hierarchy of regions. Once the tree has been constructed, we use the uniform homogeneity slice to prune the tree and filter the complex hierarchies into a single segmentation. Experimental results on NASA/JPL AIRSAR and DLR E-SAR images demonstrate the effectiveness of the proposed approach.