Image classification using ontology based improved visual words

Abstract

Multi-class classification has become a challenging task in computer vision. Due to the richness of visual concepts in the real world, the number of categories in this task is growing to large scale number of classes. Categories in multi-class data are often part of an underlying semantic taxonomy. Recent works in object classification has found it useful to use this taxonomic structure to develop more efficient recognition algorithms. In this paper, we introduce a new visual word generation and feature representation method for multi-class image classification based on semantic taxonomies. We leverage the semantic taxonomy to define visual words which are aware of contents and categories and design a hierarchical classifier based on semantic taxonomies. Experimental results show that the proposed method has improved the accuracy of classification results.