J. Zhang, J.-H. He, and Y. Fu (Eds.): CIS 2004, LNCS 3314, pp. 1175–1180, 2004.
© Springer-Verlag Berlin Heidelberg 2004
A Brushlet-Based Feature Set Applied to Texture
Classification
Tan Shan, Xiangrong Zhang, and Licheng Jiao
National Key Lab for Radar Signal Processing and Institute of Intelligent,
Information Processing, Xidian University, Xi’an, 710071, China
tanshan5989@yahoo.com.cn
Abstract. The energy measures of Brushlet coefficients are proposed as fea-
tures for texture classification, the performance of which to texture
classification is investigated through experiments on Brodatz textures. Results
indicate that the high classification accuracy can be achieved, which
outperforms widely used classification methods based on wavelet.
1 Introduction
The analysis of texture image plays an important role in image processing. Much work
has been done to develop proper representations that are effective to texture analysis
during the last several decades [1][2][3]. Recently, two spatial-frequency techniques
were introduced, namely, Gabor filters [4] and wavelet transforms [5][6], both of
which have achieved considerable success in texture classification. Especially, the
texture image analysis methods based on wavelet have received more and more atten-
tion in that the energy measures of the channels of the wavelet decomposition were
found to be very effective as features for texture analysis.
It is well known that the orientation is an important characteristic of texture.
Unfortunately, the separable 2-D wavelet transform provides only few orientations
other than horizontal, vertical and diagonal ones. Many researchers have been
dedicating to the problem, and hope to develop certain new mathematical tool which
can provides more orientation information of image than wavelet does [8][9]. In paper
[10], the author introduced a new system called brushlet, which is a new kind of
analysis tool for directional image. And the ability of brushlet to analyze and describe
textural patterns was well demonstrated by compressing richly textured images
efficiently in paper [10].
In this paper, we show how the brushlet provide efficient features for texture image
classification.
This paper is organized as follows. First, the brushlet transform is discussed briefly.
And in section 3, energy measure of brushlet coefficients as texture feature is pre-
sented. Then, in the section 4, the effectiveness of brushlet energy measure is investi-
gated experimentally. Finally, the conclusions are drawn in section 5.