A segmentation method for remote sensing image region on Riemannian manifolds

Abstract

Focus on the issue of rotation and scale in-variance for remote sensing image(RSI) segmentation, a feature extraction and classification method is proposed based on differential space. A RSI is divided into many regions with different size, and all the covariance matrices of each region are calculated. Those covariance matrices construct a connected Riemannian manifold. The map relation between the Riemannian manifold and a Tangent space is built that contains an Exponent and a Logarithmic matrices computation. Furthermore, the distance measure is established on the Riemannian manifold. It is employed to segment regions of a RSI. Experiment results show that the method is efficient and has robust rotation and scale invariance.