Output details
13 - Electrical and Electronic Engineering, Metallurgy and Materials
Queen Mary University of London : A - Electrical and Electronic engineering
Fast Multidimensional Entropy Estimation by k-d Partitioning
Entropy estimation for multidimensional distributions suffers from the 'curse of dimensionality' beyond only a few (about 5) dimensions. This paper introduces a new fast method, scaling as O(n log(n)), several orders of magnitude faster than competing methods, allowing 30 or more dimensions. Method applied to build a new image feature extractor, multi-dimensional Scale Saliency (MDSS), which 'remarkably decreases computation time' compared to previous techniques [Suau & Escolano, http://bit.ly/1eZl5Bf]. This has proved to be practical and computationally efficient, and has been applied to shape comparison [http://bit.ly/HlTTRH], active robot vision [http://bit.ly/16DXk1k], and texture categorization [http://bit.ly/16DXkyw].