During every iteration of the level sets, the shape prior has to be updated with the pose information. I have got struck in this pose estimation and so the shape is not directed towards actual object. Could anyone suggest how to update the shape information.
One approach to classifying an image using level sets is partition an image into disjoint sets and characterize each disjoint set in a partition with a level set function $\Phi_k$. Classification results from minimizing a functional with respect to $\Phi_k$ (see, for example, the attached image).
This is the approach in
J.-F. Aujol, G. Aubert, L. Blanc-Feraud, Wavelet-based level set evolution for classification of textured images, IEEE Trans. on Image Proessing 12(12), 2003:
http://www.math.u-bordeaux1.fr/~jaujol/HDR/A1.pdf
A good, detailed analysis of level sets is given in
M. Sussman, A level set approach for computing solutions to imcompressible two-phase flow, Ph.D. thesis, UCLA,1994.
http://www.math.fsu.edu/~sussman/thesis.pdf
See Section 2.3, starting on page 7.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Human-Computer Interaction