International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004 Interr 
2.5 
As a 
sense 
erode 
disap 
textu 
appe: 
have 
mate 
imag 
foun 
textu 
amot 
17 di 
show 
textu 
amot 
and | 
whic 
  
Figure 5: The scanned Medusa before and after scan post-processing Rath 
techr 
need 
2.4 Post-processing scans 
Several of the extracted models had to be post-processed. On 
the one hand, damage to the physical elements had to be 
digitally restored. On the other hand, 3D data coming from the 
‘shape-from-video’ and 'shape-from-still pipelines are not 
always of sufficient quality for close-up shots. The problematic 
3D models were manually enhanced using surface editing tools 
for polygonal objects, which are readily available. We used 
Alias’ Maya (Alias, 2004), which has excellent edit tools like 
surface smoothing, sculpting and stitching. Fig. 5 shows on the 
left raw 3D data, which first have to be stitched together. In the 
next step, the surface was smoothed to remove noise (resulting 
from the shape-from-video method) and the effects of erosion. 
The damaged snake (= big hair curl) of the Medusa was 
restored by using Maya's sculpting tool. The resulting surface 
rendered with its texture is shown in figure 5 on the right. 
2.6 
With 
put a 
Some architectural elements of the Nymphaeum (e.g. the 
decorations on the aediculae) contain so much detail (cavities) 
that their structured-light scans contain holes and yield 
enormous polygon counts (see top of fig. 6). In addition, most 
of the fine elements are damaged. While holes can be filled with 
cleanup software like Paraform (Paraform, 2004) and high 
resolution models can be reduced with e.g. Maya, a combination 
of all three aforementioned unfavourable characteristics makes 
restoring difficult: the reduce functions imply an unacceptable 
data loss and editing the cleaned high resolution surfaces is too 
time-consuming and too complicated. In addition, today's 
computer hardware is still too slow. Hence, we propose 
restoring such surfaces via depth map painting. Fig. 6 illustrates 
the process: in a first step, a depth map is rendered (for non- 
planar objects, the depth map can be extracted patch-wise with 
more sophisticated methods like the one proposed by 
Krishnamurthy and Levoy (Krishnamurthy, 1996). Then the 
depth map can be easily retouched with image manipulation 
software like Photoshop (Adobe, 2004) by using the common 
painting, drawing, and retouching tools. Finally, the restored 
depth map is converted back into a polygonal surface (or the 
depth map can be stored as bump/displacement map). For each 
of these steps several alternative software solutions are 
available. 
    
Figure 6. Restoration of the scanned decorations via retouching 
of the depth map.