A semantic-based image retrieval system using a hybrid method K-means and K-nearest-neighbor

Abstract. Semantic-based image retrieval is one of the important problems for the multimedia system and has garnered great interest in recent years. This paper proposes an improved method to build a self-balanced clustering tree, called iC-Tree. We build the Semantic-Based Image Retrieval (SBIR) based on iC-Tree named SBIR_iCT. We propose a hydrid method, which is a combination of K-mean, K-Nearest-Neighbor (KNN) to build iC-Tree, to improve the classification performance of this tree. The system uses images that are segmented into different regions and extracted visual features of each region. Semantic concepts are analyzed from those features based on the proposed C-Tree. The result of the retrieval process is a set of similar images with semantic concepts, which matches the query image and a visual word vector. hen, we design ontology for the image dataset and create the SPARQL query by extracting semantics of image. Finally, the semantic-based image retrieval on iC-Tree (SBIR_iCT) model is created hinging on our proposal. Experiments on 20,000 images of ImageCLEF dataset confirm that our proposed methods improve the retrieval performance, compared with an image retrieval system using K-means clutering, which we proposed earlier. At the same time, this performance is compared with some of recently published methods on the same dataset, show that our proposed methods are effective in handling problems of SBIR.