![]() |
Pablo
Arbelaez |
Interactive Segmentation
Collaboration with Laurent Cohen
| User Constraints | Segmentation |
![]() |
![]() |
With what accuracy can the true object boundaries be predicted, starting only from a single point inside each object?
Description
The purpose of this project is twofold. First, we propose an algorithm for constructing constrained segmentations from Ultrametric Contour Maps. This allows our technique to be used as a post-processing of any hierarchical segmentation scheme. Second, we suggest a framework for evaluating quantatively constrained segmentation algorithms. We benchmark our method and the watershed and markers approach from morphology.
Code
- Matlab code.
Results
- Browse the results on the test set of the BSDB (100 images).
- Download a single results file.
- Download the data to reproduce the results of the paper.
Publications
- P. Arbelaez and L. Cohen. Constrained Image Segmentation from Hierarchical Boundaries. To appear in CVPR 2008. [pdf]