The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
Subsequently a consistent model combining the statue and the 
niche data was generated by co-registration using corresponding 
points on the walls of the niche. 
For texture mapping the statue dataset mentioned above was 
used. Because of lack of texture information in the invisible 
regions, only a not fully photorealistically textured model can 
be presented. In fact the missing texture was filled up by 
artificial texture patterns. 
2. DATASETS 
To reconstruct the Small Buddha a number of potential data 
sources, especially in form of images, were available from the 
web, in print media or from individual persons. But also a 
contour map was available. Most of the images were useless for 
the image-based reconstruction, because of the bad image 
quality or the low resolution. From all these potential data 
sources only three were used to reconstruct the Small Buddha. 
A. The first dataset is a series of nine terrestrial images, 
acquired in August 2003, using a digital Sony Cybershot 
camera, with 6 Megapixel image format. These images have 
been used to cover the empty niche after the demolition of 
the statue. 
Figure 2. Contour line map 
(courtesy Prof. Minoru Inaba ) 
B. The second dataset consisted of two amateur images, 
acquired using a 5.5 x 5.5 cm 2 analogue YASHICA camera 
with a focal length of approximately 80 mm, used 1975 by 
Hans Werner Mohm, covering large parts of the statue 
before demolition, see Figure 1. This dataset is 
characterised by very oblique viewing directions and a 
small base-distance ratio. In addition, the camera 
parameters, except a rough approximation of the focal 
length, were unknown. The advantage of these images is the 
fairly large image scale and therefore an adequate small 
object pixel size. 
Figure 1. Images of the dataset B 
C. The third dataset is a contour map, published in (Higuchi, 
2001), and generated by using metric images of a 
photogrammetric campaign, which are not available to the 
public, see Figure 2. The quality and resolution of the 
printed map, especially in steep regions, is insufficient, due 
to merging lines. In some regions it is not possible to 
separate them from each other. Furthermore, the 
equidistance and the absolute heights were unknown. 
3. DATA PROCESSING SEQUENCE 
The workflow for the extraction of 3D information from the 
three datasets and the combination of these data to generate a 
full 3D model of the Small Buddha, is shown in Figure 3. 
Figure 3. Workflow of the model reconstruction procedures 
4. PROCESSING OF THE INDIVIDUAL DATASETS 
4.1 Dataset A 
The orientation of the niche dataset, including the estimation of 
the camera parameters was done in a previous work with a 
standard photogrammetric bundle adjustment using SGAP, an 
in-house developed software. Ground control points, acquired