ng systems 
vards rapid 
ind texture 
pping data 
cessing of 
ose-ranged 
1gly. Then, 
juity of the 
inalysis. 
| automatic 
"automatic 
parameters 
uildings is 
ing facades 
lamely the 
resented to 
d with the 
t of known 
> is finally 
osaic point 
geometric 
completed 
ontributing 
ON 
affected by 
eliminated 
mates lens 
s thanks to 
istortion is 
  
o" 
     
  
of 
n 
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
eliminated by applying parameter Æ, only parameter Æ is 
estimated in the article. 
As Fig.la illustrated, a straight line is extracted in the image 
while the actual line is slightly curved due to lens distortion. 
Therefore the line extracted is dimidiated and each line section 
is extracted again. The extracted result is illustrated as Fig.1b. 
The two lines extracted cannot obviously satisfy the condition 
of collinearity thanks to lens distortion. As Fig.lc illustrated, d 
denotes the displacement of point C to point Cl caused by lens 
distortion. Therefore, d can be employed to calibrate the lens 
distortion. A model of least adjustment is then evolved 
according to this conclusion to estimate the radial lens 
distortion parameter £1. 
3. OCCLUSION REMOVAL 
Since the occlusion of trees goes against the extraction of lines 
from buildings, the occlusion area should be automatically 
distinguished and removed. 
3.1 Occlusion Removal Employed Hue Information 
As Fig. 2a illustrated, it is taken for granted that color 
information can be employed to remove the occlusion of trees 
because of the color similarity of trees. Therefore, a *80—200' 
range of hue value is chosen to determine whether a pixel lies 
in the occlusion area. The result of occlusion removal is 
illustrated in Fig.2b. As we can see, some pixels in the red 
ellipse area which hue values are out of the range we chosen are 
remained although they are lying in the occlusion area, 
meanwhile some pixels in the green ellipse area vice versa. A 
conclusion is drawn, thereupon, that the result of occlusion 
removal only employed hue information comes out lacking 
stability. 
Here it is presented to improve the stability that an algorithm of 
Occlusion removal employed information of hue and lines. 
3.2 Occlusion Removal Employed Information of Hue and 
Lines 
Lines are Firstly extracted in the original image as Fig. 2a 
  
   
¢. Occlusion removal employed 
information of hue and lines 
b. Occlusion removal 
employed hue information 
Fig.2 Occlusion removal 
illustrated. As we can see, lines on the building façade are 
almost corresponding to object lines parallel to X-axis and Y- 
axis in the object space, therefore their orientation are regular 
while lines in the trees vice versa. The information of lines with 
regular orientation, thereupon, can be used to detect the 
occlusion area combined with hue information. The process is 
presented below: 
1) Divide the original image into 40x40 rectangular grids 
with grid size of 18 pixelsx12 pixels. 
2) Count the number of pixels which hue values are in the 
range of 80-200 in every grid. If the number is 80% of 
total, this grid is deemed to lie in occlusion area and 
removed. Otherwise, the grid will be quartered and the 
same operation will impose on new grids. 
3) The density of lines with regular orientation in the whole 
image is determined by Eqs. 
y Li 
Pr = Sr (D 
With, 
p; : The density of lines with regular orientation in the 
whole image; 
Li : The length of line i(pixel); 
5: The area of the whole image(pixel). 
4) Calculate the densities of lines with regular orientation 
respectively in grids having not been removed. If the 
density of a grid is smaller than that of whole image, this 
grid is deemed to lie in occlusion area and removed. 
The result of occlusion removal, as illustrated in Fig.2c, is 
obviously better than that illustrated in Fig.2b. 
4. THE AUTOMATIC RECTIFICATION OF FACADE 
TEXTURE 
Since building 
facades are planar, 
there. are» two 
groups of lines 
parallel to X-axis 
and Y-axis in 
object space 
respectively on 
each of them. As 
a result, only 
focus / and three angular orientation parameters ¢ ,® ,k can 
Do: Xol lyo 
  
Fig. 3 The process of rectification 
be calculated 
according to 
theories of 
vanishing point 
geometry. Therefore 
facade textures are 
rectified employed f 
and ¢ , ® , k with 
    
principle point fixed X 
at the center of 
image. The process 
Z 
Fig. 4 the relationship between p; 
and object space coordinate system