ibul 2004 
  
s in order 
re several 
'ovement. 
;, missing 
petween a 
1e texture 
oy details 
stence of 
jean they 
tree or a 
ual model 
sions may 
point is 
texture of 
y is when 
in happen 
resulting 
example, 
camera to 
nstrument 
vered 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
Figure 4. An occluded building. 
from the neighbouring stations, which have a smaller 
vertical angle to the missing part. A texture example is shown 
in Figure 5 where the instrument could not capture the top part 
of the building face; thus its texture is incomplete and is shown 
using white colour. The third situation is when a building face 
is too long or too high; thus using a single station may cause 
features in the resulting texture to appear stretched, as the 
images are captured with high perspectives. Figure 6 shows an 
example of such, where features are stretched especially at the 
top. 
Te 
Figure 5. Texture with a missing part. 
439 
    
Figure 6. High perspective effect. 
In most of the above situations, it is desired to improve the 
quality of the texture by filling in the gaps or integrating better 
quality images into the affected arcas. Unfortunately, there is no 
system capable of dealing with such problems easily and 
efficiently. Some techniques simply fill in the affected part with 
an artificial pattern or an image-based template. Applying such 
methods, however, does not lead to a truly realistic solution to 
the problem, as the improved section may still be different from 
the way it looks in reality. 
One possible solution is to take original images in a way that 
occlusion does not occur in the resulting textures. For this, each 
image is examined on site and is replaced with a new one if 
necessary. Most current digital cameras allow for reviewing 
images immediately after they are captured. Thus if unwanted 
details are included or if the desired features are not present, the 
relevant images are simply deleted and new ones taken. Taking 
images in such a way, however, requires further on-site 
supervision and interaction by the user. In addition, this method 
may not be practical in busy parts of cities. 
Fortunately, here, as the terrestrial images are all registered to 
the same coordinate system and the texture processing is carried 
out based on object points, the textures of a specific face can 
easily be merged. This means, if parts of a texture are affected, 
they can be replaced by textures processed using images from 
the neighbouring stations. 
As a first step, the face containing the affected area and the 
stations that cover this area are defined by the user. Using 
ATPT, a low resolution texture for the affected arca is created 
from each individual station. The textures are then examined 
visually and the one with the best view over the area of interest 
is selected. Then, a high resolution texture, having the same 
spacing interval as that of the final texture, is created using the 
images from the selected station. 
The unaffected parts, taken from the original texture, and the 
affected part taken from the texture just created are fused to 
form a new texture. Indeed, this texture is the result of merging 
the two texture parts. As both textures have been created at the 
same resolution and belong to the same object (ie. the 
corresponding face), their dimensions and orientation are the