Hyperspectral image classification by recursive spatial boosting based on the bootstrap method

We consider contextual classification of hyperspectral data based on the boosting method. Bootstrap AdaBoost proposed by Kawaguchi and Nishii (2006) is applied to Spatial Boosting for contextual classification. The paper proposes a recursive version of Spatial Boosting. Posterior probabilities of each pixel are updated by the contextual classification function derived from Spatial Boosting and this is repeated. The proposed method with random stumps shows excellent performance for classification of AVIRIS data. Furthermore, it is superior to other well-known contextual classification methods including MRF-based classifiers.