A Reliable Image Retrieval System Based on Spatial Disposition Graph Matching

E. Zagrouba(1*), S. Ouni(2), W. Barhoumi(3)

(1) Higher Institute of Computer sciences, Unité de Recherche en Systèmes Intelligents en Imagerie et Vision Artificielle (URSIIVA-ISI), Tunisia
(2) Faculty of Sciences of Tunis, Unité de Recherche en Systèmes Intelligents en Imagerie et Vision Artificielle (URSIIVA-ISI), Tunisia
(3) Higher School of Technology and Computer Sciences, Unité de Recherche en Systèmes Intelligents en Imagerie et Vision Artificielle (URSIIVA-ISI), Tunisia
(*) Corresponding author


DOI's assignment:
the author of the article can submit here a request for assignment of a DOI number to this resource!
Cost of the service: euros 10,00 (for a DOI)

Abstract


Content-based retrieval of multimedia information is one of the most difficult research topics in multimedia computing and information retrieval. However, the domain of image retrieval is still an open problem in comparison with the obtained results in the domain of text retrieval. In this paper, we present a reliable approach for image indexing and retrieval mainly based on the spatial positions of significant regions extracted from the studied images, in order to reduce the semantic gap between the query image and the retrieved ones. For modelling spatial configuration, we propose a novel graph-based data structure that we called Partial Graph of Relative Spatial Relations (PGRSR). The proposed graph can be considered as a better candidate for modelling semantically rich data. It permits to describe the degree of relative positioning between the significant regions not necessary adjacent. Thus, the image retrieval will consist principally to the resolution of a graph matching problem. For this, we present a novel algorithm of graph matching while proposing a measure of similarity between graphs allowing the ranking of the retrieved images. Besides, we combine both text and image features for image indexing and retrieval in order to more reduce the semantic gap. The efficiency of the proposed approach is illustrated while studying many scenarios on two different databases. The first database is homogenous and contains 120 images (composed of 560 regions) of countries flags. The second database contains 600 heterogeneous images (composed of 3220 regions) of diverse real-world indoor and outdoor scenes. The promising recorded results were evaluated while using the precision and recall assessment criteria.
Copyright © 2018 Praise Worthy Prize - All rights reserved.

Keywords


Graph of Relative Spatial Relations; Image Indexing and Retrieval; Matching; Regions Segmentation; Textual Keywords

Full Text:

PDF


Refbacks

  • There are currently no refbacks.



Please send any question about this web site to info@praiseworthyprize.com
Copyright © 2005-2019 Praise Worthy Prize