Texture image retrieval based on directional filter selection and statistical modeling
Abstract
This paper presents a new approach to texture image retrieval based on the statistical modeling in the multidirectional complex transform domain for effectively improving the performance of the retrieval system. Firstly the basic directional filters are selected and used for the complex transform. Then the coefficients of highpass subband are modeled as generalized Gaussian distribution; the magnitude coefficients of complex directional bandpass subbands are modeled as Gamma distribution; and their distribution parameters are efficiently combined with the statistical features of relative phases of complex bandpass subband coefficients. Finally the texture images of VisTex standard database are retrieved using a new combined form of similarity measurements. The experimental results show that, in comparison with the existing texture image retrieval methods, the proposed approach obtains a superior average retrieval rate.
- Publication:
-
Tenth International Conference on Graphics and Image Processing (ICGIP 2018)
- Pub Date:
- May 2019
- DOI:
- 10.1117/12.2524174
- Bibcode:
- 2019SPIE11069E..23Q