Content-based image retrieval based on iRS-Tree and ontology

Abstract

Content-based image retrieval is implemented based on low-level features and applied to numerous fields. However, the major challenge of this approach is the semantic gap between low-level features and high-level concepts. Therefore, the problem of semantic-based image retrieval is attractive to improve query accuracy. In this paper, a semantic-based image retrieval model using the improved R^S-Tree structure (iR^S-Tree) and ontology was proposed. The improvements on iR^S-Tree include: (1) improving the operation of adding elements to reduce tree building time; (2) improving the operation of splitting nodes to enhance clustering accuracy. The result of querying images on iR^S-Tree is a set of similar images and a set of visual vocabulary. Then, a SPARQL command is automatically generated from this visual vocabulary and queries a semi-automatic ontology to extract semantics for the image. Experiments were performed on three image data sets: COREL, Oxford Flower-102, and CUB-200-2011. Experimental results were evaluated and compared with recent works on the same data set to demonstrate the effectiveness and correctness of the proposed method.