A hypergraph reduction algorithm for joint segmentation and classification of satellite image content

Abstract

In this paper, we introduce a novel hypergraph reduction algorithm, and we evaluate it in an innovative method for joint segmentation and classification of satellite image content. It operates in 3 steps. First, we compute an Image Neighborhood Hypergraph representation (INH). Second, we reduce the INH model and we exploit a morphism from INH to Reduced INH (RINH) to generate superpixels. Then, we perform a superpixels supervised classification according to their features. Our approach is very fast and can deal with great sized images. Its reliability has been tested on several satellite images with comparison to single pixelwise classification. © 2010 Springer-Verlag.