Superpixel-based change detection in high resolution sar images using region covariance features

Feature representation is very important for high resolution synthetic aperture radar (SAR) image interpretation, especially for unsupervised change detection. In this paper we propose a superpixel-based change detection approach that utilize region covariance as feature representation. After segmenting SAR images into superpixels, the second order statistic of SAR feature vectors, i.e., the region covariance feature is extracted for each superpixel. In the difference map generation stage, the dissimilarities of corresponding superpixel pairs in multitemporal SAR images are measured by calculating the Bartlett distances between region covariance features. After that, an adaptive thresholding method is applied to obtain the final detection results. Two multi-temporal TerraSAR-X high resolution SAR image sets are tested for the proposed approach and promising results are achieved.