Nuclear and other posters wall for SC99

Computational Forensics

Problem: Identification of skeletal remains

Project goal: Create recognizable (identifiable) facial reconstruction working from CAT/MRI of unknown skull

The Process
The unknown skull is matched against a database of known skulls to build a patchwork skull "twin".
Then the flesh/skin data from the known skulls is added to the patchwork skull to create a patchwork reconstruction.
The patchwork reconstruction is then morphed to add features (such as eye color) and make it suitable for rotation.

Traditional methods:

Clay reconstruction built on the skull
2-D Photographic reconstruction

Advantages to computational reconstruction:

minimize handling of skull
3-d reconstruction can be rotated (enhancing recognition)
reconstruction can be delivered via web interface
multiple versions (e.g., different eye colors) can be generated

MRI of researcher with 24 key points marked

Geometric mask with scalp area removed and holes for eyes

Mask morphed to the 24 points with eyes added

Photo of researcher

Challenge	Solution
MRI images do not show hair. This artificial baldness hinders recognition. Putting in simulated hair that is not correct also hinders recognition.	The scalp portion of the image is removed, letting the user mentally fill in an appropriate hairline.
MRI image surfaces have many, many triangles, making the image surface rough and time consuming to render and rotate.	There are 24 points which are key to facial recognition. These points are captured from the MRI and mapped onto a smooth geometric mask. The mask is then morphed to match the 24 points.

ORNL || CSM || CCS || CPED || DSRD || LSD

URL http://www.csm.ornl.gov/SC99/IAO/Miscwall.html
v2-11/3/99 webmaster