Fast Correspondence-Based System for 2-D Shape Classification

In earlier work, we developed a fast method for 2-D shape retrieval based on point correspondences of silhouette contours. That method did not assume that the database shapes had class labels. If class label information is added, then the shape retrieval system can be used for 2-D shape classification. This paper explores this application, by extending our prior shape retrieval method to make use of class labels, and then evaluating its classification accuracy on a large test database containing 70 object classes. The results show that the method is both highly accurate (96.5-99.6%) and fast (<1 second).

• Type: Conference
• Year: 2003
• Where: CISST
• Authors: Boaz J. Super
• Keywords: 2-D Shape, CISST 2003, Class Labels, Hardware, Shape Retrieval