View-Independent Face Recognition with Hierarchical Mixture of Experts Using Global Eigenspaces
Reza Ebrahimpour, Farzad M. Jafarlou · Journal of Communication and Computer · 2010
A new tree-structured and efficient parallel learning algorithm for view-independent face recognition applications that is so-called ”hierarchical mixture of experts” is presented. Problem space in structure of hierarchical mixture of experts (HME) is divided into several subspaces for the mixture of experts and these subspaces are split into several new subspaces for experts. Consequently, the outputs of the experts and mixture of experts are combined with mediation of a gating network. The experimental results show that computational complexity is reduced, and error rate 12.51% is reduced in comparison with conventional MEs.