Image retrieval by categorization using LVQ network with wavelet domain perceptual features

  • Authors:
  • M. K. Bashar;Noboru Ohnishi;Kiyoshi Agusa

  • Affiliations:
  • Graduate School of Information Science, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, Japan;Graduate School of Information Science, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, Japan;Graduate School of Information Science, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, Japan

  • Venue:
  • PCM'04 Proceedings of the 5th Pacific Rim Conference on Advances in Multimedia Information Processing - Volume Part II
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

Though most textile images are pattern dominant, there found a limited researches that focus on the pattern characteristics. In this study, we propose some perceptual features (directionality, regularity, symmetry) in the wavelet domain. Correlation among wavelet coefficients is the basis of the above features. In order to reduce searching time, we first categorize the database using supervised LVQ network. For each class, a class-vector is formed through averaging all the feature vectors in that class. The query key is first compared with class-vectors to come up with a category. It then performs similarity comparisons with the population of the selected category and retrieves relevant images. Users have also the provision to interact with the system if query fails to capture the relevant class. An experiment with a set of 300 curtain images shows the effectiveness of the proposed features compared to the well-known Gabor or discrete wavelet energy signatures.