Related Experiment Videos
Recognition of shapes by editing their shock graphs
Thomas B Sebastian1, Philip N Klein, Benjamin B Kimia
1GE Global Research Center, PO Box 8 KWC 218A, Schenectady, NY 12301, USA. sebastia@crd.ge.com
IEEE Transactions on Pattern Analysis and Machine Intelligence
|October 6, 2004
Summary
This study introduces a new shape recognition framework using deformation distances. It achieves 100% accuracy in top matches for object recognition, proving effective for silhouette-based applications.
Area of Science:
- Computer Vision
- Pattern Recognition
- Computational Geometry
Background:
- Object recognition is crucial in computer vision.
- Traditional methods struggle with shape variations and high-dimensional data.
- Silhouette-based recognition requires robust shape matching techniques.
Purpose of the Study:
- To develop a novel framework for object recognition using silhouette shape matching.
- To address the challenge of high-dimensional deformation spaces in shape analysis.
- To create a computationally efficient and accurate shape recognition system.
Main Methods:
- Measuring shape distance via minimum deformation.
- Employing shock-graph topology and transitions for shape equivalence classes.
- Utilizing an edit-distance algorithm on shock graphs for optimal deformation path finding.
Main Results:
- The framework provides intuitive shape correspondences.
- It demonstrates robustness against various visual transformations.
- Achieved 100% recognition rates within the top three matches on two databases.
Conclusions:
- The proposed framework offers a practical and effective solution for silhouette-based object recognition.
- The shock-graph approach significantly reduces computational complexity.
- The method shows high potential for diverse shape-based recognition applications.