Mining visual web knowledge utilizing multiple classifier architecture

Abstract

Inspite of the huge amounts of image data on the web, mining image data from the web is paid less attention than mining text data, since treating the semantics of images is much more difficult. This paper introduces a new system to mine visual knowledge on the web that aims to build a Domain Oriented Image Directory by using the Earth Mover's Distance and Color signatures. Instead of using a flat classifier to combine text and image classification, the system suggests dividing the classification task into smaller classification problems corresponding to the branches in the classification hierarchy. Thus a multiple classifier system is presented. This paper illustrates the suggested system and discusses each of its components. Extensive experiments were conducted to test the system and also to compare it with commercial search engines. By the experiments we show that the proposed system accuracy outperforms the mostly used commercial search engines. © 2010 IEEE.