TEXTURAL ANALYSIS FOR REMOTELY SENSED IMAGERY 
Xinghe Sun" 
* China University of Geosciences, 
Ping Qin' " 
Beijing 100083 
* * Commission for Intergrated Survey of Natural Resources 
Chinese Academy of Sciences, 
ABSTRACT 
According to defined the texture as the measurement of 
image gradient of gray, two parameters can be obtained 
to describe remote sensing image quantitatively. Which 
are the texture intensity and the texture density. The tex- 
ture intensity is used to show the differences ef one pixel 
with its neighbours, while the texture density is used to 
show the frequency of changes of gray level. 
In addition the relative gradient are defined to describe 
the texture characteristic of image from the relative 
changes of gray level. The producted texture images 
from this method include both the information of space 
and part of information of spectrum. So the texture 
structures in the some over bright and dark areas can be 
shown obviously. 
In this paper, we also discuss how to select the size of 
the moving windows. 
INTRODUCTION 
The texture extraction techniques is important for the 
image analysis. So far, there are various approaches to 
Rosenfeld.A (1975), Haralick. R. 
N(1979) have thoroughly discussed and generalized 
texture extraction. 
these approaches to different types of imagery which in- 
cluded aircraft or satellite multispectral scanner images 
and microscopic images of cell culture or tissue samples. 
etc. 
This paper concentrates on the approaches to texture 
analysis of remotely sensed imagery. Giving the texture 
of remotely sensed imagery a specific definition, it in- 
troduces the concepts of texture intensity and texture 
density, and their corresponding extraction techniques. 
From the differences and gradient of gray level between 
neighboring pixels, the generated images of texture in- 
tensity and texture density reflect properties of the orig- 
inal image's texture. The paper also introduces the rela- 
tive gradient approach, which uses the variation of gray 
level between pixels to signify the structure of tiny tex- 
58 
Beijing 
ture, to enhance the coarse texture of the original image 
and to display the structure of texture within the highly 
bright and dark areas. 
THE PROPERTIES OF TEXTURE 
OF THE REMOTELY SENSED IMAGERY 
The remotely sensed imagery records the information of 
electromagnetic waves radiated from the earth surface. 
The variation of gray levels within the imagery is re- 
flected by the spectral features of different objects on 
the earth surface. While the structure of the image tex- 
ture is decided by difference between different gray lev. 
els. The structure shows the distribution of radiation 
from the earth which relates to time and location. So the 
texture structure of the image is random and irregular. 
Based on these properties, we define the texture of 
remotely sensed imagery as "measurement of the rate of 
spatial variation of gray levels within the image". The 
textural statistical value I is defined as 
dD 
I= ff (—) D 
UE ( 
where dD is difference of gray levels between neighbor- 
ing pixels, dL is distance between the pixels. 
Based on this definition, two measurements are pro- 
posed to characterize textural structure of image. They 
are, 
e Texture intensity gradient of gray level varia- 
tion of pixels in the window. It discribe the differ- 
ence gray level between neighboring pixels of 
image. 
e Texture density distribution of gradient of gray 
levels of pixels in a two dimensional area. It 
discribe frequency of variation of the pixel gray 
level. 
The two measurements are mutually independent, but 
they complement each other 
in characterizing the 
textural features. Since tone and texture are interrelated 
concepts, calculating the value of relative gray level va- 
riation of pixels, i.e. the relative gradient, can be used to 
ext 
Or 
dL 
te: 
If 
T! 
en 
ar 
te 
=