Multiscale Segmentation of High Resolution Satellite Imagery by Hierarchical Aggregation

A multiscale segmentation approach by hierarchical aggregation(SHA) is proposed,which is inspired by the idea of SWA(segmentation by weighted aggregation).SHA is more suitable for high resolution satellite imagery through some improvement over SWA.The imagery is first preprocessed by watershed transformation so that a graph is constructed from the segments by their spectral distance.Then a bottom-up aggregation framework derived from algebraic multigrid solver that is used to minimize a graph cut measure.Region based features(e.g.texture,shape,etc.) are accumulated conveniently during the aggregation process and implementation also is faster than SWA.