You are here
3D Skeletonization /Volume Thinning
Thinning is the process of reducing a shape to a simpler version that still retains the essential features of the original object. The thinned version of the shape is called the skeleton. When the skeleton is centered with respect to the boundaries of the original object, it is referred to as the medial-axis or medial-surface. Skeletons of 3D shapes have various applications
- Fast Shape matching and Tracking
- Automatic navigation
- Shape abstraction
- Animation Control
The complexity of the desired skeletal structure depends on the application. Tracking and matching applications require the skeleton shape to accurately model the original object. Such a skeleton is very dense since it has to capture every small feature in the original object. The skeleton can be used as the path to automatically navigate through 3D objects. If the skeleton is centered, collisions with the object boundaries are avoided. Such a skeleton used for navigation can be called a centerline. It is required to be very thin, ideally 1 voxel wide. It is useful for a single algorithm to address these diverse requirements and generate skeletons suited for either application. We have developed an algorithm to control the density of the skeleton by a single parameter called the thinness parameter.
MPEG Movies
- Automatic Navigation (Virtual Endoscopy) through the Human Colon. (MPEG. 1.1MB)
- Automatic Navigation through the Human Trachea. (MPEG. 650KB)
- Volumetric Animation of an insect. (MPEG 475KB)
- Visible Human Animations. Link
Examples of Skeletons
1. Skeleton of a cuboidal and ellipsoidal shape.
2. Effect of the Thinness Parameter. Note that a higher thinness parameter results in a thinner skeleton. The sharp curvature discontinuity at the cylinder faces is modeled by an increasing number of spines in the skeleton.
3. Skeletons in Shape Matching and Tracking
Since the skeleton retains shape information and is a reduced representation, it can be used to match shapes. The matching process is faster because of the fewer number of points in the skeleton. The figure below shows vortex structures from a turbulence simulation. The original vortices shown on the left have been thinned to about 12% of the original as seen in the right image. The thinned version can then be used for fast feature tracking.
Table 1 : Speedup Using Skeletons for Feature Tracking
Thinness | %Error | Skel. Time(sec) | Tracking Time (sec) | %Total Time |
None
|
0
|
0
|
437.0
|
100
|
0.5
|
4.32
|
314.64
|
109.25
|
97
|
1.0
|
5.15
|
300.23
|
69.30
|
84.56
|
2.0
|
14.50
|
272.08
|
22.45
|
67.40
|
Nikhil Gagvani