Cross-shape Cluster Extraction by Extended Linear Fuzzy Clustering

Abstract

Linear clustering is a promising technique for extracting intrinsic local linear sub-structures. This paper proposes a new approach for extracting cross-shape clusters formed by two lines, in which two lines identification is jointed into one process of finding a single linear sub-structure by rotating prototypes. By rotating one of the two lines around intersection, crossing prototypes are formulated only by a single basis vector. In order to solve the size problems caused by prototypical lines with infinite length, the combined objective function with FCM is also considered.