This paper presents a hybrid learning machine for human identification. It is a merger of eigenface with fisherface method, genetic fuzzy clustering and complex neural network. The non-linear aggregation based summation and radial basis function neural networks (NLA-SRBF NNs) are proposed as one of the functional component of the novel learning machine. The architecture of NLA-SRBF NNs incorporates hidden neurons, with summation and radial basis aggregation, and output neurons with only summation aggregation, along with complex resilient propagation (ČRPROP) learning procedure. The improved learning and speedy convergence of NLA-SRBF NN enables the hybrid machine to provide better recognition accuracy. The learning machine consists of feature extraction, unsupervised clustering and supervised classification module. The aim of our proposal is to enhance the performance of biometric based recognition system. The efficacy and potency of our hybrid learning machine demonstrated on three benchmark biometric datasets-extended Cohn-Kanade, FERET and AR face datasets to comprehend the motivation. The performance comparisons of different variations of hidden neuron and learning algorithm thoroughly presented the superiority of the proposed NN based hybrid learning machine.

This paper illustrates the new structure of artificial neuron based on root-power means (RPM) for quaternionic-valued signals and also presented an efficient learning process of neural networks with quaternionic-valued root-power means neurons (ℍ-RPMN). The main aim of this neuron is to present the potential capability of a nonlinear aggregation operation on the quaternionic-valued signals in neuron cell. A wide spectrum of aggregation ability of RPM in between minima and maxima has a beautiful property of changing its degree of compensation in the natural way which emulates the various existing neuron models as its special cases. Further, the quaternionic resilient propagation algorithm (ℍ-RPROP) with error-dependent weight backtracking step significantly accelerates the training speed and exhibits better approximation accuracy. The wide spectrums of benchmark problems are considered to evaluate the performance of proposed quaternionic root-power mean neuron with ℍ-RPROP learning algorithm.