Improving the Efficiency of Semantic Image Retrieval Using a Combined Graph and SOM Model

Abstract

Extracting similarity and semantic images from images is a hot topic that is used in various semantic retrieval systems. The GP-Tree, a hierarchical clustering tree, is used in the article to retrieve semantic images. The GP-Tree is then used to create a graph of neighboring leaf nodes, resulting in a smaller search list and a reduced chance of missing related similar images. In addition, the neighbor graph is used to build a Self-Organizing Map (SOM) to improve clustering efficiency and image retrieval performance. To improve high-level semantic retrieval efficacy, we also propose extending the current ontology structure. The ontology framework with rich domains represents most of the basic objects. With this ontology, the sets of images are added for the enrichment and can be used for many different sets of images. The SPARQL query is automatically generated from visual words for querying on the ontology. The query result on the ontology is a set of similar images and the definition of its semantics. The proposed method is tested on the WANG and ImageCLEF datasets, and the results are compared to previous publications on the same dataset, demonstrating its efficacy.

Keywords

• Semantic-based image retrieval
• content-based image retrieval
• hierarchical clustering
• self-organizing map
• ontology