295 
extraction; Using the variational spectrum knowledge: it uses 
the various normalization indices mainly, and these indices 
adopt the ratio transformation and the normalization treatment 
mainly, and the values are between -1 and +1. It makes ground 
feature get the maximal brightness enhancement on the index 
image, and the other background ground features get the 
common restraint. The indices in common use are vegetation 
index, wave index , building index etc. 
The method of discovering the spectrum knowledge includes 
typical sampling method, spectrum curve method and spectrum 
section plane method[3]. The information extraction based on 
spectrum knowledge, need to be diacritical between ground 
feature and background on the spectrum, and there is less same 
spectrum phenomenon between ground feature and background, 
and the internal spectrum of ground feature should be consistent. 
It can carry out extraction according to the ground feature 
internal characteristic component spectrum when the internal 
spectrum of the ground feature disaccords. It must carry out 
extraction according to the other knowledge of the ground 
feature when there is more same spectrum phenomenon 
between the ground feature’s internal characteristic component 
spectrum and the background’s. 
2.2 Remote Sensing Image Information Extraction Based 
on Texture Knowledge 
The image texture has reflected the character and spatial 
relation of the image grayscale. The texture is a kind of spatial 
distribution that the adjoining pixel gradation obeys some kind 
of statistical array in the certain image area. The texture 
includes the structure texture and the non-structure texture. 
When the ground features’ composing are complicated and they 
are bigger than the sensor’s spatial resolution, their structure 
and the composing may be sensed remotely, and there is the 
obvious texture characteristic on the image. When it is difficult 
to resolve information extraction’s problem completely 
depending on spectrum knowledge, and when there is texture 
structure characteristic of differentiating the background, it 
must together use the ground feature’s spectrum knowledge and 
texture knowledge in information extraction. 
The method of discovering texture knowledge includes gray 
level co-occurrence matrix method, half variation function 
method, fractal and fractal dimension method, Markov random 
field and wavelet transform method, extremum texture method, 
structural unit method etc, gray level co-occurrence matrix 
method and half variation function method are in common use 
among them. 
Gray Level Co-occurrence Matrix: It’s definition is that if one 
area of an image have N gray scales, the gray level co 
occurrence matrix of the corresponding area is a N X N matrix, 
and the definition of the pixel on (i, j) is that the pixel who’s 
grayscale is i moved a given distance (displacement d) meets 
the pixel who’s grayscale is j, the probability this condition 
appears is p (i , j) [3] . It can calculate 13 kinds of texture 
measures from the gray level co-occurrence matrix, they are: 1) 
Angular Second Moment (Energy); 2) Contrast; 3) Correlation; 
4) Variance; 5) Homogeneity; 6) Mean; 7) Standard Deviation; 
8) Entropy; 9) Dissimilarity. 
Half Variation Function: It is the geostatistics fundamental 
implement. It can look remote sensing image’s grayscale as a 
regionalization variable, and every pixel can be considered as 
the realization of regionalization variable, it has not only 
randomness but also spatial structure. According to the 
definition of half variation function, the value of remote sensing 
image’s half variation function reflects both the pixels’ 
structure and the image data’s statistics characteristic at the 
certain degree, and this is the demand of describing texture 
characteristic method, therefore it is reasonable to describe 
remote sensing image texture with half variation function [4] . 
Half variation function category: 1) direct half variation 
function; 2) absolute half variation function; 3) alternate half 
variation function; 4) false alternate half variation function; 5) 
weighting half variation function 
2.3 Remote Sensing Image Information Extraction Based 
on Shape Knowledge 
It is not only same or close on the spectrum characteristic but 
also similar on the texture characteristic sometimes between the 
ground feature and the background. It carries out information 
extraction according to the ground feature’s shape knowledge in 
this condition. 
2.3.1 Remote Sensing Image Information Extraction 
Based on Area Shape Knowledge: The ground feature 
displays the certain shape on the remote sensing image. The top 
or side shape of the ground feature is showed on the image, the 
ground feature has the proper shape generally, for example, the 
road and the river are line-type, the building is regular etc. The 
ground feature also displays the certain size on the image, so it 
can differentiate according to the size for two kinds of ground 
features form which are difficult to differentiate owing to the 
similar shape. 
The factors which express the shape knowledge are the area, the 
perimeter, the length and the width. The methods of discovering 
the ground feature’s shape knowledge are the method based on 
the perimeter and area, the method based on the area and the 
method based on the area and length of the region. 
The method based on the perimeter and area. E.g. 
k=4aTp (d 
The method based on the area. Compactness index. E.g. 
C 4 =J7TA c ,C i =A/A c (2) 
The method based on the area and length of the region. Aspect 
Ratio. E.g. 
FR = Al L 1 . (3) 
Ellipse index. E.g. 
EI = [tt(ì/2L)]L/A. (4) 
where A = the area of the ground object