Taking a Break from the Archaeology Research...

I found this week's reading ("Reanimating the Dead: Reconstruction of Expressive Faces from Skull Data" by Kahler, et al.) to be very interesting. Being a DMDer, these ACM papers are more my style. Yeah, yeah...I'll admit that I skipped over the majority of Section 4.2, which talks about the math involved in the procedure. But something about this research just pulled me in. Maybe it's because my senior project could potentially add hair to the reconstructed faces? Something else worth noting is the similarities of this "art form" with those we discussed in class (like Pate's paintings). "The process is thus highly dependent on rules of thumb, the experience of the artist, and some guesswork," says the paper. The same thing could be said about Pate's work too.

The paper discusses a technique of computerized facial reconstruction using what is refered to as the "tissue depth method." This involves using statistical tissue thickness measurements at specific points on the face to guide how the face is modeled on top of the skull. In this case, the facial "modeling" consists of deforming a template mesh head. For non-computerized versions of the method, the facial "modeling" consists of applying and shaping clay layers to a cast of the skull. There are several advantages of using Kahler's computerized method. First of all, it is a lot faster than the non-computerized alternative (taking about an hour to do vs. "many hundreds"). Secondly, the computerized reconstruction process makes it possible to easily create many "variants" for a single skull, the most common being related to the build of the reconstructed person - skinny, medium, or obese. Finally, the results produced by Kahler (et al) allow for simple facial animations (such as smiles or frowns) to be applied to the reconstructed faces, which "could be helpful for identification purposes." The paper focuses on applying its facial reconstruction technique to the field of forensic science and criminology. However, I definitely see its application for digital "peopling" of reconstructed scenes/architecture of the past.

Here's the general step-by-step approach used by Kahler (et al) to reconstruct a face, given the skull of the person and details about where it was found and what it was found with:

1. In cooperation with an anthropologist, extrapolate more information from the remains, including an estimation of age, ancestry, sex, and stature.
2. Scan the skull (using a volume or range scanner) to obtain a 3d representation (a triangle mesh). The original data should be simplified as little as possible.
3. Equip the skull model with landmarks (which look like the "dowel" tissue depth markers used by tranditional tissue depth reconstructors). Each landmark is associated with a vector in surface normal direction and is scaled to local tissue thickness based on statistical data.
4. Deform a template head model according to the landmark data. This model consists of several "layers", including the skin surface and virtual muscles right below it. A mass-spring system connects the layers together in a realistic fashion, which comes in handy for animating simple expressions later on.
5. Add more landmarks (if necessary) in places where tissue thickness is near-constant. This is accomplished by interpolating existing landmarks in terms of position and thickness.
6. Apply empirical heuristics about the shape relations between skin and skull to the model (for example, rules governing the width of the nose, length of the nose, width of the mouth, thickness of the lips, etc). Update the landmark configuration accordingly.
7. Apply expressions (smile or frown) to the face by adjusting the virtual muscles, which then pull on the skin layer according to the mass-spring relationship between the two layers.
8. Add colors (or textures) to the skin, lips, and eyebrows of the face model.
   

Here are the layers of the template head model, as described in 4 (I thought this was one of the coolest parts of Kahler's technique):