Nonlinear Adaptive Manifold Self-Organizing Map with Reproducing Kernels and its Application to Pose Invariant Face Recognition

概要

論文の詳細を見る
Adaptive Manifold Self-Organizing Map (AMSOM) is an evolution of Self-Organizing Maps, where each computational unit defines a linear affine subspace (manifold). The affine subspace in a unit is represented by a mean vector and a set of basis vectors. After training, these units result in a set of affine subspace detectors. In complex situations, however, these are not enough to describe a class of patterns because of its linearity. In this paper, the AMSOM in the high-dimensional space with reproducing kernels, referred to as Nonlinear Adaptive Manifold Self-Organizing Map (NAMSOM) is proposed in order to achieve efficient classification and visualization of relations between classes. By using the reproducing kernels, linear affine subspaces in the AMSOM can be extended to nonlinear affine subspaces easily. To verify the effectiveness of the proposed method, it was applied to face recognition under varying pose as a practical example.
社団法人電気学会の論文
2005-06-01