New support vector-based design method for binary hierarchical classifiers for multi-class classification problems

Abstract

We propose a new hierarchical design method, weighted support vector (WSV) k-means clustering, to design a binary hierarchical classification structure. This method automatically selects the classes to be separated at each node in the hierarchy, and allows visualization of clusters of high-dimensional support vector data; no prior hierarchical designs address this. At each node in the hierarchy, we use an SVRDM (support vector representation and discrimination machine) classifier, which offers generalization and good rejection of unseen false objects (rejection is not achieved with the standard SVMs). We give the basis and new insight into why a Gaussian kernel provides good rejection. Recognition and rejection test results on a real IR (infrared) database show that our proposed method outperforms the standard one-vs-rest methods and the use of standard SVM classifiers.