Multi-dimensional complex-valued Gabor wavelet networks.

This paper proposes a new complex-valued wavelet network based on Gabor transfer functions which have complex-valued weights, input/output variables and translation parameters. A learning algorithm is developed by using the technique of genetic algorithm to facilitate implementation of the network. The effectiveness of the proposed network is evaluated by applying the network to approximate several complicated nonlinear functions, and to tackle the 4-QAM (quadrature amplitude modulation) digital communications channel equalization and the image enhancement via feature extraction problems in signal processing. Indeed, the evaluation results obtained by testing the proposed wavelet network with real facial and textile fabric images clearly show that it is a very useful tool for solving problems in the field of complex-valued signal analysis.