Template-based fuzzy clustering with cluster-wise coordinate transformation
Katsuhiro Honda, Shotaro Osaka, Akira Notsu, Hidetomo Ichihashi · 2012
FCM-type clustering algorithms have been extended to various shape recognition models with non-point prototypes such as lines, quadric curves, etc. The template-based clustering model is a modified FCM-type algorithm, in which arbitrary shape cluster prototypes are constructed based on template data sets. The clustering criterion between a data point and a template is defined by searching the nearest point from a template data set in each cluster. In this paper, a new approach for constructing cluster prototypes is proposed by introducing cluster-wise coordinate transformation of template structures. Optimality evaluation by template matching is also applied in order to avoiding local optimality of rotation.