Discovering multi-sense features from images using random indexing

Abstract

Random Indexing is a recent technique for dimensionality reduction while creating Word Space model from a given text. The present work explores the possible application of Random Indexing in discovering feature semantics from image data. The features appearing in the image database are plotted onto a multi-dimensional Feature Space using Random Indexing. The geometric distance between features is used as an indicative of their contextual similarity. Clustering by Committee method is used to aggregate similar features. In this paper, we show that the Feature Space model based on Random Indexing can be used effectively to constellate similar features. The proposed clustering approach has been applied to the Corel databases and motivating results have obtained. © 2014 Springer International Publishing.