International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXX V, Part B3. Istanbul 2004 
  
control points are obtained from the 3m resolution GIS-data. The 
field of interest in the open landscape is mostly flat, and wind 
erosion obstacles are always higher. Thus, 3D information as 
shown in figure 3 is also one of the most useful sources of 
information, which can be integrated into a combined model 
together with GIS-data and color information to support the 
extraction of wind erosion obstacles. 
3. EXTRACTION OF WIND EROSION OBSTACLES 
3.1 Image segmentation 
Image segmentation by NDVI value for CIR images to extract 
vegetations is a well-known approach (c.f. Lyon 1998, Butenuth 
2003). NDVI calculations are based on the principle that actively 
' growing green plants strongly absorb radiation in the visible 
region of the spectrum such as Red region while strongly 
reflecting radiation in the Near Infrared region and thus a high 
NDVI value (NDVI = (NIR-Red)/(NIR+Red)). There are other 
segmentation approaches based on Vegetation Index. They all 
give the similar results (Lyon 1998). So NDVI is used to segment 
the images in this paper. Sometimes, results of segmentation with 
NDVI are not very satisfying due to noises. The results have to be 
improved with other approaches. 
In 1931, Commission Internationale de l'Eclairage (CIE) 
presented a device independent color space CIE XYZ, as shown 
in figure 4 (SEII EM-MI 2002). Generally, the points of surface 
X+Y+Z=1 are considered in the three-dimensional area. 
This surface includes the white point (Wx,Wy, Wz) with value 
(0.312779, 0.329184, 0.358037) and three settlements for 
the basic colors R, G, B. Each point in color space RGB has its 
corresponding point in color space CIE XYZ. 
Z (blue) 
    
Y (green) 
7 
| White paint 
a point of color 
Tree 
^ K (red) 
Fig. 4. Definition of CIE XYZ 
We will simply give the transformation from RGB to CIE XYZ, 
please see (SEII EM-MI 2002) for more detail of how to derive 
the transformation model. 
X 0.412291 0.357664 0.180209| | R 
Y |=| 0.212588 0.715329 0.072084 |e| G Ch 
Z 0.019326 0.119221 0.949102] | B 
786 
The color space CIE L*a*b is announced in 1976 (SEIT EM-MI 
2002). The model CIE L*a*b refers a little to the models of colors 
expressed by the Newton's circle, as shown in figure 5. 
white max luminance 
    
L = Lightness 
black zero luminance 
Fig. 5. Definition of CIE L*a*b 
The component L represents the light Lightness with value from 0 
to 100 as defined below. 
L-116-(Y/Wy)^ -16 if 0.008856 <(Y/Wy) 
(2) 
L = 903.3-(Y/Wy) else 
The components a and b represent two differences defined below. 
In theory, component a varies from blue (with value —120) to red 
(with value +120), component b varies from green (with value 
—120) to yellow (with value +120). 
a =500-(F(X/Ws)=F(Y/Wy)) 
b - 200-(F(Y/Wy)- F(Z/Wz)) = 
Where (X.Y.2) is the point to be converted, (Wx, Wy, Wz) 
the white point as the definition of CIE XYZ. The point 
(X, Y.) can be obtained from equation (1) with RGB value of 
a certain pixel in the image. F(p) — p^ if p: 0.008856, 
otherwise F(p)- 7.787 * p? «16/116. 
There is no direct relation between RGB and CIE L*a*b. The 
transformation from RGB to CIE L*a*b must be made indirectly 
through CIE XYZ. Firstly, RGB is transformed into CIE XYZ 
with equation (1), and then the received points into CIE L*a*b 
according to equation (2) and equation (3). 
The component a of CIE L*a*b will be always positive for 
vegetations in the CIR imagery, and close to +120 for the 
strongest vegetative growth. So this information can be combined 
with NDVI to segment the CIR images. Figure 6 shows the 
segmented CIR image by NDVI and CIE L*a*b information. In 
order to keep all potential wind erosion obstacles, we adopt a 
relatively low threshold during segmenting. White areas in the 
Su 
no 
m: 
lez 
sei 
im 
alg 
are