Goal :
Synthesis 3D stone models from scanned models by
“Statistical Learning of Multi-Dimensional Textures”
(master thesis of Ziv Bar-Joseph)
Algorithm
:
1.
Scanning
a few stones on 3D scanner.
After
converting we have VRML 2.0 models (scaned_stones)
2. Transforming 3D models into 2D signals by mapping
it onto
spherical domain, cubic domain:
and by transferring the cubic domain to image:
3.
Creating “Steerable Pyramid” ( part 3.3
of thesis ) for each
input 2D signal,
that is trees represents spectral defragmentation
of 2D signal.
4 Filters
that used in creating Steerable Pyramid
was taken from following link :
ftp.cis.upenn.edu/pub/eero/steerpyr.tar.gz
5. Generating a new random tree by
“statistically merging” ( part 5.2 of thesis )
the trees of the input signals.
6. Creating 2D signal from generated (
merged ) tree.
7. Transforming the image domain to cubic,
and
then transforming 2D signal back into 3D model.
8. Mapping
texture on 3D model.
In addition, there is an option to synthesize
the textures by the same process ( steps 2-5 ).