Local relational string and mutual matching for image retrieval

  • Authors:
  • Adel Hafiane;Bertrand Zavidovique

  • Affiliations:
  • Laboratoire Vision et Robotique ENSI de Bourges, Université d'Orléans, 88 Boulevard Lahitolle, 18020 Bourges, France and Institut d'ELectronique Fondamentale, Bítiment 220, Universi ...;Laboratoire Vision et Robotique ENSI de Bourges, Université d'Orléans, 88 Boulevard Lahitolle, 18020 Bourges, France and Institut d'ELectronique Fondamentale, Bítiment 220, Universi ...

  • Venue:
  • Information Processing and Management: an International Journal
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Exhibiting new features and likely related matching techniques to efficiently retrieve images from databases remains an open problem. This paper is first devoted to such a novel description of coloured textures by LRS (local relational string) so based on relative relations between neighbour pixels and on their distribution. It is illumination-invariant and able to capture some semantics local to objects through selected image stripes. Second, we propose a bi-directional query/candidate matching based on region - i.e. potential object - mutual preferences. The remaining third of the paper reports on the experimentation over 60,000 images, providing results in the TrecEval format about the performance compared with common available techniques and about the influence of key parameters in the proposed segmentation and pairing processes.