A supervised discriminant subspaces-based ensemble learning for binary classification

Abstract

A technique, fairly similar to the Karhunen-Love transform, is the Fukunaga-Koontz transform (FKT); this effective discrimination technique can be utilized where two-class classification issues crop up enabling retrieval of second-order relations of Gaussian distributed data [1, 2]. Implementation of this technique has led to diverse results such as identification and ensuing tracking of targets [3, 4], recognition of faces [2], and obtaining hyperspectral images [5]. Perusal of existing work in the field over the past few decades reveals that kernel-based learning techniques have often been employed to extract statistics of non-Gaussian data. For example, in the work of Liu et al. the kernel version of FKT (KFKT or kernel FKT) has been used to prove the presence of small targets in forward-looking infrared images [6].