Automatic building detection from satellite images using stacked generalization architecture

This paper proposes an automated segmentation, based algorithm for building detection in satellite images. The proposed method consists of a two layer hierarchical classification mechanism. In the first layer, each segment is classified by N different classifier according to different features. In the second layer of the mechanism, the class membership values of the segment from different first layer classifiers are concatenated to form a new vector, which is used by the meta classifier to classify the selected segment. The paper also presents the performance results of the proposed model and comparison with the single layer classifiers.