Ontology based classification for multi-label image annotation

Abstract

Image annotation has been an important task for visual information retrieval. It usually involves a multi-class multi-label classification problem. To solve this problem, many researches have been conducted during last two decades, although most of the proposed methods rely on the training data with the ground truth. To prepare such a ground truth is an expensive and laborious task that cannot be easily scaled, and “semantic gaps” between low-level visual features and high-level semantics still remain. In this paper, we propose a novel approach, ontology based supervised learning for multi-label image annotation, where classifiers' training is conducted using easily gathered Web data. Moreover, it takes advantage of both low-level visual features and high-level semantic information of given images. Experimental results using 0.507 million Web images database show effectiveness of the proposed framework over existing method.