A new connectionist approach for facial identification
Paul G. Luebbers, A.S. Pandya, R. Sudhakar · 2002
A new artificial neural network architecture called Power Net (PWRNET) and Orthogonal Power Net (OPWRNET) has been developed. Based on the Taylor series expansion of the hyperbolic tangent function, this novel architecture can approximate multi-input multilayer feedforward perceptrons, while requiring only a single layer of hidden nodes. This allows a compact representation with only one layer of hidden layer weights. The resulting trained network can be expressed as a polynomial function of the input nodes. The degree of nonlinearity of the network can be controlled directly by adjusting the number of hidden layer nodes, thus avoiding problems of over-fitting which restrict generalization. The OPWRNET architecture was applied to the task of facial image recognition. An architecture of one and two hidden layer nodes were trained and compared to a linear discriminator. Features were extracted from the images using normalized centralized regular moments. The extracted moments were combined into individual features for each order of the moment by generating receptive fields for each feature.>