Level Set Segmentation with Shape Priors

1
Level Set Segmentation with Shape Priors

• Jue Wang and Jiun-Hung Chen
• CSE/EE 577 Spring 2004

2
Notes on Level Set

• Steps for implementation
• Define your own energy functional
• Initialize the curve and surface
• Evolve according to the numerical solution
• Continue until stopping criteria is reached

3
Notes on Level Set

• So the key is
• Defining a good energy function
  according to your application
• For basic image segmentation, the energy
  functional is usually defined as

4
Notes on Level Set

• Numerical Implementation
• Not easy to figure out but theyre in books, for
  example
• Numerical stability should be also considered
  when you design your own energy function

5
Segmentation with Shape Prior
Reference shape
Level set seg. w/o shape prior
Level set seg. w/ shape prior
Input image
D. Cremers and S. Soatto. A pseudo-distance for
shape priors in level set segmentation. IEEE
Workshop on Variational, Geometric and Leve Set
Methods in Computer Vision, 2003.
6
Segmentation with Shape Prior

• Basic Idea
• Shape prior can be represented as another
  implicit function
• The evolution of current is influenced by
  the distance between and .
• Problem how to define the distance

7
Segmentation with Shape Prior

• A Pseudo-distance
• Shape energy
• Evolution function

8
Segmentation with Shape Prior

• Pose parameters
• Numerical solution help needed for this ?

9
Results

• An images worth of thousands of words
• An EXE is worth more

10
Future Work

• How to incorporate statistical shape models?
• Have a bunch of reference shapes instead of one
• X.M. et al. Integrating prior shape models into
  level-set approaches. Pattern Recognition
  Letters, April 2004.