The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
704 
Blender. At the beginning, the whole dataset with 600k 
triangles was processed. With this number of triangles, the 
performance of Blender was very poor. Nevertheless, a VRML 
model was exported with the full number of triangles (600k). 
The quality was satisfying, but real time navigation was more 
or less impossible. Due to this, a reduced dataset of 100k 
triangles was generated. With this dataset, the performance of 
the real time visualization was acceptable. The comparison of 
the visual impression, especially of the visible details, showed 
that the two datasets are equivalent, because most of the 
information for the human eye is provided by the texture, not 
by the geometry. This leads to the fact, that for pure 
visualization the geometry can be reduced down to a certain 
level, as long as the full texture resolution is still used. Finally, 
a virtual flight around the object was produced with the full 
dataset and a resolution of 1024 x 768 pixels. Figure 9 shows 
the result of the Khmer head dataset. It was generated using 12 
images. 
Figure 9. Final result, the textured Khmer head 
5. CONCLUSIONS 
The presented workflow of texture mapping contains all the 
important steps, from the given orientation data and the original 
images to the final textured 3D model. It includes a new 
visibility analysis algorithm fully based on vector algebra. This 
leads to an image-resolution independent analysis, suitable to 
handle sparse and dense datasets at the same time, in a fully 
automatic way without manual interaction. The following 
Triangle to Image Assignment procedure selects the best texture 
source for each triangle from multiple images. No averaging is 
done, this preserves the high frequent texture information. The 
finest details of the images are preserved. To achieve a 
photorealistic and seamless textured model, three image 
enhancement steps were implemented. First, the vignetting was 
removed using a simple cos4 relation. A second step removed 
the global brightness difference over the whole image, caused 
for example by different exposure times. In a last step, a local 
brightness correction was conducted to compensate influences 
of different or moving light sources or spot light effects by 
generating a brightness difference surface over the whole image. 
This surface was interpolated using a biharmonic spline 
function. Common points in two images are used as input. 
The final visualization of the data was done using the open 
source software system Blender, which provides the possibility 
for high resolution rendering of single images as well as for the 
generation of movies. 
6. FURTHER WORKS 
The described algorithms are working well and achieve 
satisfying results. Nevertheless, some parts of the algorithms 
will be improved in the future. First, the performance of the 
visibility algorithm concerning processing time should be 
improved. A possible solution is the splitting of the dataset in 
tiles and usage of multi core CPU systems. 
Concerning the Triangle to Image Assignment, a patch growing 
algorithm will be implemented, to reduce the length of 
borderlines between different texture sources. This will reduce 
the number of potential seams. 
Furthermore, the algorithm of vignetting reduction can be 
enhanced in order to model more complex lens systems. 
A last improvement could be the exchange of the cross 
correlation function by a least squares matching procedure to 
improve the point fitting for images with larger angle between 
the viewing directions. 
ACKNOWLEDGEMENT 
The author thanks the Swiss National Science Foundation (SNF) 
for the financial support to make this work possible. 
REFERENCES 
Akca, D., Remondino, F., Novk, D., Hanusch, T., Schrotter, G., 
and Gruen, A., 2007. Performance evaluation of a coded 
structured light system for cultural heritage applications. 
Videometrics IX, Proc. of SPIE-IS&T Electronic Imaging, San 
Jose (California), USA, January 29-30, SPIE vol. 6491, pp. 
64910V-1-12. 
Amhar, F., 1998. The Generation of True Orthophotos Using a 
3D Building Model in Conjunction with a Conventional DTM. 
IAPRS, Vol. 32, Part 4 “GIS-Between Vision and Applications”, 
pp. 16-22. 
Biasion, A., Dequal, S., Lingua, A., 2004. A new Procedure for 
the Automatic Production of True Orthophotos. ISPRS 
Conference proceedings, Istanbul, Commission IV, pp. 538-543. 
Blender, 2008. http://blender.org , (accessed 20. Jan. 2008) 
d’Angelo, Pablo, March 21 st-24th 2007. Radiometric alignment 
and vignetting calibration, Workshop: Camera Calibration 
Methods for Computer Vision Systems - CCMVS2007, 
Bielefeld University, Germany. 
El-Hakim, S., Gonzo, L.,Picard, M., Girardi, S., Simoni, A., 
2003. Visualisation of Frescoed Surface: Buonconsiglio Castle 
- Aquila Tower, Cycle of Months. Proceeding of International 
Workshop on Visualisation and Animation of Reality-based 3D 
Models, Tarasp-Vulpera, Switzerland. 
Frueh, C., Sammon, R., Zakhor, A., 2004. Automated Texture 
Mapping of 3D City Models with Oblique Aerial Imagery. 2nd 
International Symposium on 3D Data Processing, Visualisation 
and Transmission (3DPVT’04), Thessaloniki, Greece, pp 275- 
282.