Extracting an ontology of portrayable objects from Wordnet

Abstract

We describe building a large-scale image ontology using the WordNet lexical resources.\nThis ontology is based on English words identifying portrayable objects. We reviewed the\nupper structure and interconnections of WordNet and selected only the branches leading\nto portrayable objects. This article explains our pruning approach to WordNet. The\nwords, which represent portrayable objects, are then used as queries in our VIKA (VIsual\nKAtaloguer) system which acquires images through a web image search engine, performs\ncontent-based image indexing and clustering. Coherent images form clusters and others\nare rejected. So images inside clusters mostly represent the object determined by the\nquery, and in this way image collections representing objects are created. An ontology of\nportrayable objects with representative images in its nodes will be a useful tool for\nsolving the object recognition task.