Skeleton and elliptic shape models
The goal of this project is to improve image segmentation quality in cases when low-level features such as class-dependent color models or contrast edges alone do not provide enough information. Primary objectives of this project are:
- Build a model of an image that combines low-level features with a high-level model of object shape, which, in turn, can describe object classes with great in-class variety of shape.
- Develop an efficient image segmentation algorithm for the model.
- Develop an algorithm for learning shape models from images augmented with pixel-wise labelings or, perhaps, a weaker kind of ground truth.
We currently focus on a class of graph-based shape models inspired by the medial axis transform: meaningful object parts are represented by edges of the graph and an additional value representing object width is assigned to every graph vertex. Energy function defined on graphs with same structure is used to control shape variety.
List of relevant papers
- B. Yangel, D. Vetrov. Learning a Model for Shape-Constrained Image Segmentation from Weakly Labeled Data. In Proceedings of International Workhop on Energy Minimization Methods (EMMCVPR2013) , 2013. pdf
- B. Yangel and D. Vetrov. Image Segmentation with a Shape Prior Based on Simplified Skeleton. Proceedings of International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR) , LNCS 6819, 2011. pdf
Have you spotted a typo?
Highlight it, click Ctrl+Enter and send us a message. Thank you for your help!