The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
464 
2.1 Texture Feature in Classification of TM/ETM+ Image 
Although the ETM+ image covers a large area and is suitable 
for resource investigation, the number of bands is too limited to 
distinguish some landcovers, which share a similar spectral 
feature. Through a plenty of experiments, we discovered that it 
is effective to use texture feature to enhance the objects' feature 
so as to be separated from others. For example, entropy and 
variance are used to extract lakes from ETM+ (Armenakis, 
2003), and edge density assisted by morphological operation is 
very helpful to find the city boundary in TM images (Hlavka, 
1987). However, these textures are ineffective to extract 
urbanized area in ETM+ of Beijing metropolis. 
2.2 Definition of Skewness and its Current Application 
Skewness, in digital image, describes the asymmetrical attribute 
of a group of pixel values. If the skewness is equal to 0, it 
means the data is symmetrical rigorously, and the farther from 0, 
the more asymmetrical. For a gray scale image, (x,,x,,...,Jt n ) 
represents a set for all pixel values in a moving window, so 
skewness is defined as a kind of texture as follows: 
, I 
skewness = -= 
{n-\){Vf 12 
(i) 
Where n, M, V represent the number, mean value and variance 
of gray scale values, respectively in a moving window. 
As an important statistical measure, skewness is widely used in 
finance, engineering and atmosphere research. As the come 
forth of remote sensing, skewness is becoming used in such 
field, though not widely. Skewness can be used to detect and 
enhance the edges in the image (Irons, 1981), however the 
author pointed out that he does not find this feature can be used 
for landcover classification. And skewness is also a tool to 
detect small targets in images [Chiang, 2001]. Compared to 
optical remote sensing, skewness is so widely used in radar data 
process that ERDAS has a skewness function in the RADAR 
modular. 
2.3 Feasibility of Skewness in Extracting Urbanized Area 
the separability, a low pass filter was used to reduce the 
intra-distance of patches, so we got a good separability (Figure 
3c, 3f, 4c). 
(c) (<i) 
(e) (f) 
Figure 3. (a) True color ETM+ Image of crop field A (b) 
skewness value of crop field A (moving window in 9*9 size) 
(c) low pass operation on Figure 3b (d) True color ETM+ 
Image of urbanized area B (e) skewness value of urbanized 
area B (moving window in 9*9 size) (f) low pass operation on 
Figure 3e 
For urbanized area, experiencing frequent reconstruction, it is 
composed by buildings, vegetation, streets, water body, so its 
symmetry is much less than that of crop field. Since skewness 
can reflect the symmetry of different landcover, it is an effective 
and simple feature to distinguish urbanized area and crop field. 
As the result of spectral analysis, there are various kinds of 
crops around urbanized area of Beijing, and some of them 
(Figure 3a) have very similar spectral value to some urbanized 
area (Figure 3d). So, if they are extracted directly by spectral 
feature (Figure 4a), they will be misclassified into urban in a 
high probability. 
Calculating the skewness value from the panchromatic band, 
Figure 3b, 3e were generated from Figure 3a, 3d. A 
computation on crop field in Figure 3b and urbanized area in 
Figure 3e show that the average skewness in former one is 
much lower than the latter one. But these two landcovers are not 
well separable (Figure 4b) due to big intra-distance in both two 
landcovers, though their inter-distance is obvious. To enhance 
Figure 4(a) Average albedo from crop field A and urbanized 
area B in the ETM+ image