An improvement of R-Tree for content-based image retrieval

Abstract. The problem of image retrieval from a large image dataset has been having many challenges. Therefore, indexing and searching images on a data structure are important requirements to improve retrieval effectiveness. In this paper, an improved structure based on R-Tree, named \(R^S\)-Tree (Region Sphere Tree), is proposed to enhance the accuracy of the content-based image retrieval on different image sets. The improvements on \(R^S\)-Tree include:
(1) representing image feature vectors in the form of spheres to optimize storage space;
(2) improving operations on \(R^S\)-Tree such as adding, deleting, and splitting nodes to enhance retrieval efficiency. The result of this processes creates the balanced clustering tree. Since then, content-based image retrieval model is built rely on \(R^S\)-Tree. On the base of proposed theory,the experiment is performed on data-sets including COREL, Wang, Oxford Flowers-17 with precision values of 76.29%, 73.16%, and 78.69%, respectively. The experimental results are compared to related works on the same dataset to demonstrate the effectiveness of the image retrieval model based on \(R^S\)-Tree.