Clustering of Learning Images and Generation of Multiple Prototypes for Object Recognition

common features in all learning objects only. The In this paper, we propose two methods of clustering learning images to generate prototypes automatically for object recognition. One is for clustering views of a single object and the other is for clustering different objects observed in a similar direction which belong to a same object class. In both two cases, we first group all learning images into several clusters by considering appearance features and spatial relations between these features, then we construct a prototype in each cluster. generated model which contains common features of all learning objects can be regarded as a prototype. However, most objects, even if they belong to a same object class, can not be described by only one prototype because of a large variety of appearance features and spatial relations among the features. In order to solve the problem of how many prototypes are needed to "cover" all learning objects, we may divide all learning objects into...

• Type: Conference
• Year: 1998
• Where: MVA
• Authors: Jin Jia, Keiichi Abe
• Keywords: Appearance Features, Common Features, Learning Objects, MVA 1998, MVA 2007