Multi-dimensional scale saliency feature extraction based on entropic graphs

Abstract

In this paper we present a multi-dimensional version of the Kadir and Brady scale saliency feature extractor, based on Entropic Graphs and Rényi alpha-entropy estimation. The original Kadir and Brady algorithm is conditioned by the curse of dimensionality when estimating entropy from multi-dimensional data like RGB intensity values. Our approach naturally allows to increase dimensionality, being its computation time slightly affected by the number of dimensions. Our computation time experiments, based on hyperspectral images composed of 31 bands, demonstrate that our approach can be applied to computer vision fields, i.e. hyperspectral or satellite imaging, that can not be solved by means of the original algorithm. © 2008 Springer Berlin Heidelberg.