| Paper: | TP-P4.1 |
| Session: | Image and Video Segmentation V |
| Time: | Tuesday, September 18, 14:30 - 17:10 |
| Presentation: |
Poster
|
| Title: |
SHAPE PRIORS BY KERNEL DENSITY MODELING OF PCA RESIDUAL STRUCTURE |
| Authors: |
J. P. Lewis; Stanford University | | |
| | Iman Mostafavi; University of California, San Diego | | |
| | Gina Sosinsky; University of California, San Diego | | |
| | Maryanne Martone; University of California, San Diego | | |
| | Ruth West; University of California, San Diego | | |
| Abstract: |
PCA is often used for shape prior modeling, but captures only second order moment statistics. Kernel densities can reproduce arbitrary statistics, but are problematic for high-dimensional data such as shapes. An evident approach uses PCA to reduce the problem dimensionality, followed by kernel density modeling of the PCA coefficients. We show that useful algorithmic and editing operations can be formulated in terms of this simple approach, and illustrate this in the context of point distribution shape models. Particular points can be evaluated as being plausible or outliers, and a plausible shape can be completed given limited manually guided operator input. This "PCA+KD" approach is simple, scalable, and provides improved modeling power. |
