Ensemble of tensor classifiers based on the higher-order singular value decomposition

Abstract

In this paper we present an ensemble composed of classifiers operating with multi-dimensional data. Classification is performed in tensor spaces spanned by the basis obtained from the Higher-Order Singular Value Decomposition of the pattern tensors. These showed superior results when processing multi-dimensional data, such as sequences of images. However, multi-dimensionality leads to excessive computational requirements. The proposed method alleviates this problem, first by partitioning the input dataset, and then by feeding each partition into a separate tensor classifiers of the ensemble. Despite the computational advantages, also accuracy of the ensemble showed to be higher compared to a single classifier case. The method was tested in the context of object recognition in computer vision. In the paper we discuss also methods of input image prefiltering in order to increase accuracy. The conducted experiments show high efficacy of the proposed solution. © 2012 Springer-Verlag.