Ontology-based indexing of annotated images using semantic DNA and vector space model

Abstract

The study presented in this paper focuses on the preprocessing stage of image retrieval by proposing an ontology-based indexing approach which captures the meaning of image annotations by extracting the semantic importance of the words in them. The indexing algorithm is based on the classic vector-space model that is adapted by employing index weighting and a word sense disambiguation. It uses sets of Semantic DNA, extracted from a lexical ontology, to represent the images in a vector space. As discussed in the paper, the use of Semantic DNA in text-based image retrieval aims to overcome some of the major drawbacks of well known traditional approaches such as ‘bags of words’ and term frequency-(TF) based indexing. The proposed approach is evaluated by comparing the indexing achieved using the proposed semantic algorithm with results obtained using a traditional TF-based indexing in vector space model (VSM) with singular value decomposition (SVD) technique. The experimental results show that the proposed ontology-based approach generates a better-quality index which captures the conceptual meaning of the image annotations.