Object Contour Shape Modeling using a Cyclic Self-Organizing Map

Abstract

Shape modeling is an important task for object recognition，detection，and deformation．In recent years，machine learning methods are significantly developed that find secret regularities and potential variables in given data．This paper proposes shape representation of object contours by using a cyclic Self-Organizing Map that is one of the unsupervised learning method．It is difficult for a traditional cyclic SOM method to represent odd-shaped object contours．We address this problem by using the order of points located on the order of SOM units．We also propose methods of extracting shape features from object contours．Our method is examined by the experiments of silhouette recognition．