文档介绍：TEXTURE CLASSIFICATION USING WORKS AND DISCRETE
WAVELET TRANSFORM
Paul Schumacher Jun Zhang
Mayo Foundation Dept. of Electrical Engineering 8~ Computer Science
200 First Street . University of Wisconsin - Milwaukee
Rochester, MN 55905 Milwaukee. WI 53201
schumacher.******@
ABSTRACT various resolution levels can give insight into the type
of texture and, hence, aid in the definition of classifi-
This paper describes a method for classifying textured
cation boundaries.
images using works and discrete wavelet trans-
In this work, the discrete wavelet transform is taken
form (DWT). In this method. a multiresolution anal-
on textured images and used as a preprocessing tool for
ysis is applied to textured images to extract a set of
work classifiers. Four works are
intelligible features. These extracted features, in the
trained at a given resolution level, each with inputs
form of DWT coefficient matrices, are used as inputs
taken from one of the four separable DWT coefficient
to four different multilayer perceptron (MLP) neural
submatrices found at that level: low-low (LL), LH, HL,
networks and classified.
or HH. To take advantage of the quasi-periodic nature
Generalization performance is improved when a lo-
of the textures, a locally-connected, weight-sharing neu-
cally conne