On the Rayleigh nature of Gabor filter outputs

Sitaram Bhagavathy, J. Tesic and B.S. Manjunath

Dept. of Electrical and Computer Engineering
University of California at Santa Barbara
Santa Barbara, CA 93106
Email: {sitaram,jelena,manj} @ece.ucsb.edu

Abstract

Texture has been recognized as an important visual prim-itive in image analysis. A widely used texture descrip-tor, which is part of the MPEG-7 standard, is that com-puted using multiscale Gabor filters. The high dimen-sionality and computational complexity of this descrip-tor adversely affect the efficiency of content-based re-trieval systems. We propose a modified texture descrip-tor that has comparable performance, but with nearly half the dimensionality and less computational expense. This gain is based on a claim that the distribution of (absolute values of) filter outputs have a strong ten-dency to be Rayleigh. Experimental results show that the dimensionality can be reduced by almost 50%, with a tradeoff of less than 3% on the error rate. Further-more, it is easy to compute the new feature using the old one, without having to repeat the computationally expensive filtering step. We also propose a new normal-ization method that improves similarity retrieval and indexing efficiency.
[PDF] [BibTex]
S. Bhagavathy, J. Tešic and B. S. Manjunath,
IEEE International Conference on Image Processing (ICIP), vol. 3, pp. 745-748, Barcelona, Spain, Sep. 2003.
Node ID: 351 , DB ID: 149 , VRLID: 116 , Lab: VRL , Target: Conference
Subject: [Image Texture] « Look up more