REMOTE SENSING IMAGE TEXTURE ANALYSIS AND CLASSIFICATION
WITH WAVELET TRANSFORM
Changqing Zhu
Zhengzhou Institute of Surveying and Mapping
Zhengzhou 450052 CHINA
KEY WORDS: Remote-Sensing , Analysis, Modeling , Texture.
ABSTRACT:
One difficulty of texture analysis in the past was the lack of adequate tools to characterize different
scales of textures effectively. Recent develmpments in multiresolution analysis such as the Cabor and
wavelet transforms help to overcome this difficult. This paper intruducs a new approach to characterize
texture at multiple scales. The performance of wavelet transform are measured in terms of sensitivety and
selectivity for the classification of twenty remote sensing textures. The reliability exhibited by texture sig-
natures of such transforms are beneficial for accomplishing segmention, classification and subtle discrimi-
nation of texture.
1. INTRODUCTION
Textures provide important characteristics for
the analysis of many types of images including nat-
ural scenes, remote sensing data and biomedical
modalities. The perception of texture is believed to
play an important role in the human visual system
for recognition and interpretation. Perious methods
of analysis for accomplishing texture classification
maybe roughly divided into three categories : statis-
tical, structural and spectral [1][2][3]. These
methods have been successful for many fields, but
they share one common weakness. That is, the pri-
marily focus on the coupling between image pixels
on a single scale. More recently, the methods
based on multichannal or multiresolution analysis
have received a lot of attention. Recent develop-
ments in spatial/frequency analysis such as Gabor
transform, Wigner distribution, DCT , and wavelet
transform provide good mutiresolution analytical
tools. Specifically, wavelet transform plaies an im-
portant part in texture analysis.
There were some studies for texture classifica-
tion by wavelet tramform. Carter [4] first reported
texture classification results using Morlet and Mex-
ican hat wavelets. He achieved 98 percent accuracy
on 6 types of natural textures. Andrew and Jian
[5] stuied texture classification by wavelet packet
signatures. They achieved more than 98 percent ac-
curacy on 25 types of natural textures. In this pa-
per, we introuduced a new approach to characterize
remote sensing texture images at multiple-scales
with wavelet transform. The performance of
wavelet transform are measured for the classifica-
tion of twenty aerial remote sensing texture im-
ages. Wavelet representions for twenty images in
the same resolution were classified with few errors
by a simple minimun-distance classifier. The classi-
fication for six images in different resalutions had
been done, too.
2. THE METHOD OF CLASSIFICATION
2. 1 Image Features by Orthonoral Wavelet Trans-
form
By orthonoral wavelet transform, we mean the
decomposition. of an. image into multiple levels
framework. Each level and each portion represent
themselves special properties of frequency and spa-
tial. Fig. 1 shows a level wavelet decomposition, in
which c represent low frequency information, dl
the vertical edge information, d2 the horizonal edge
information, and d3 inclining edge information.
Fig. 2 shows a 4-levels wavelet decomposition and
obtains seventeen sub-images include original im-
age. Here wavelet is selected for Daubechics
1036
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
wavelet [
2. 2 Imag
T we
ages wer
metry an
clude de:
and so o.
180dp. €
and 230c
8bit/pixe
down in
wheih tw
twenty f
2. 3 Feat
Eac
teen sub
transfor!
sub-imag
Here H
where
and CO.
sample :
features
qo
texture
minimut
3. d. Clas
Sup
sample