i:bul 2004 
: part 
ee can be 
tures that 
:h (a) has 
est. This 
rts of the 
it can be 
ind has a 
ture was 
are taken 
a better 
vo steps. 
the tree 
and paint 
ice and at 
almost a 
quality of 
n the left 
t due to 
thus be 
icing. the 
>xture, on 
tched. 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
  
  
(b) 
Figure 9: Improved textures. 
To reduce the perspective distortion of the windows, a window 
in the middle of the texture was copied as the source and pasted 
over the stretched windows as shown in 10-(b). It should be 
noted that the safety of this method in terms of "reality" 
441 
  
Figure 10. Improving geometric quality of textures 
depends on how similar the sample and the source features are. 
In other words, this method can only be used if features that are 
copied have the same size, shape, and orientation. 
The above examples show how easily the textures can be 
merged in order to create more complete and realistic results. 
The process is convenient and is advantageous over the 
techniques which simply replace the affected area with a colour 
or a pattern. It should be noted that, despite being simple and 
easy to use, the merging technique does not guarantee the 
removal of all unwanted features or the filling in of all the gaps 
present in a texture. The success of this method depends mainly 
on the availability of suitable images on the neighbouring 
stations. 
Since images are captured at different times and position, 
however, it is quite likely that they form ideal textures when 
used in conjunction with each other. More importantly, as the 
textures of a given face are all of the same size, shape, and 
orientation the merging process is carried out as simply as 
possible. In other words, the convenience of the procedure is 
basically due to the appropriate structure of image acquisition 
and texture processing strategies employed in this project. 
4. CONCLUSIONS 
This paper reviewed and examined the components of a system 
developed for fast production of geometrically correct textures 
for building visually realistic models of buildings. The key 
concept incorporated in the system is the establishment of an 
indexed database of oriented images to allow automatic 
correspondence between images and the geometric model of 
buildings. This allows finding, rectifying, and mosaicing image 
portions covering a given face of a building. In addition as 
textures are all referenced to the actual object space, they can 
easily be compared and merged if necessary. This in turn, can 
lead to textures free of occlusions. 
As mentioned before, the ATPT uses a geometric model (NFR) 
to provide a starting point for creating textures. The source for 
this model can vary, provided that it leads to satisfactory 
textures. This means not only aerial photographs, which are