Eva Schiffer

Building Efficient, Accurate Character Skins from Examples
by Alex Mohr and Michael Gleicher, SIGGRAPH 2003 .

Summary:

This paper introduced a system for creating more flexible and correct linear blend skins from much more complicated examples. Since the system allows the user to create their original examples in any way they like, creating well rigged characters is far easier for artists. Since traditional linear blend skins are already in use in real-time applications, such as games, this technique will have many uses in the real world.

Problem:

How do you best represent a complex skin in a simple and easy to compute way?
How can you transform any system of rigging into a linear blend skin while still preserving most skin behavior?
How can you avoid degenerate twisting and bending artifacts?


Method:

The basic method taken in this paper was to use a system of examples gathered from a more complexly rigged skin to create a less complex, but still approximately correct linear blend skin. Additional joints were introduced into the skeleton of the linear blend skin to combat twisting or bending collapses and to given correct bulging effects for muscles on joint deformations. The method does require a large number of input examples, but since the examples are discarded after the new skin is computed, the method does not require much storage space. While the system allows user tweaking, the fact that it can be done in an automated manner makes it far more versatile. 

Key Ideas:

More complex skins can be represented approximately as linear blend skins.

Twisting and bending collapses can be combated automatically by adding additional joints to the existing skeleton.

Computing a linear blend skin from examples does not need to take an overly restrictive amount of time.

Contributions:

The process of automatically generating a linear blend skin from examples.

The process of automatically adding joints which can combat traditional twist and bend collapses in linear blend skins. 

Questions:

I would have liked to see some of the pictures in this paper in better detail, since I don't think the small print out that I have shows some of the differences very well. I also would have liked to see some of the cases where this technique can not automatically match the example skeletons as well.