A non-linear K-means algorithm and its application to unsupervised clustering
Yong Qing Yu, Alain Trouvé · 2003
A new partition criterion for pairwise clustering is proposed naturally in the probabilistic analysis framework. Its connection to the normal K-means algorithm is explained in two different views which also builds its relationship with the kernel approach introduced by Vapnik. Both synthetic examples and the challenging task of planar shape analysis have been given to show its efficiency in unsupervised pairwise clustering application.