The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
689 
This results in differences in colour and contrast on different 
sides of the building model and it becomes necessary to reduce 
these differences in order to improve the overall visual effect of 
the textured building model. Colour balancing of different 
close-range images can be done by histogram matching using 
statistical methods (Gonzales and Woods, 2002) to reduce the 
difference. It is given by, 
Bal Im=[°" 8 Im- 0ri *y/]x + R 'V 
where Bal Im = colour balanced image 
° ng Im = original image 
° ng p= statistical mean of original image 
0ng (p= standard deviation of original image 
Ref p = statistical mean of reference image 
Ref o=standard deviation of reference image 
Colour balancing needs to be carried out for each pixel on the 
respective colour bands. 
3. RESULTS AND DISCUSSIONS 
The test data set consist of building models extracted from a 
stereo-pair of IKONOS satellite imagery using SOCET and the 
respective building model tops were draped with the IKONOS 
image as illustrated in Figure 3. A series of close-range ground 
images were acquired using a few commercially available 
cameras ranging from Digital Single Lens Reflex cameras 
(DSLR) to simple Point and Shoot (PNS) digital cameras. The 
images were registered to obtain the extrinsic and intrinsic 
parameters using the methods described in the previous section 
and texture mapping was carried out. 
Figure 3. Test building models with pseudo façade textures 
Figure 4 shows a projected façade highlighted by the red 
boundary on a digital photograph pertaining to one of the 
building models used in our test data set. The resulting 
extracted façade is shown in Figure 5. The left image of Figure 
5 is rendered from a high resolution mesh whilst the right image 
is rendered from a low resolution mesh. It can be observed that 
the geometric distortions caused by using a skewed façade 
texture were not fully corrected on the low resolution mesh. 
Some aliasing effects were introduced into the image and these 
were especially evident at the seam lines between mesh 
triangles. These distortions are a by-product of resampling 
during OpenGL texture mapping transformations. There are 
many techniques to alleviate the aliasing effect and the method 
of using a finer mesh as shown in the left of Figure 5, is able to 
negate such errors and extract a geometrically corrected façade 
texture. 
Figure 4. Image with a single façade highlighted for texture 
mapping 
Figure 5. Geometrically corrected facade texture from Figure 4. 
High resolution mesh (left). Low resolution mesh (right). 
Figure 6. Test building with multiple facade textures 
Figure 6 shows the test building which has been textured using 
geometrically corrected and colour balanced façade textures. 
This particular building model required 3 digital photographs to 
provide enough views to texture the entire building model. As