Learning Vector Quantisation based recognition of offline handwritten signatures
International Journal of Biometrics
| Hi-index | 0.00 |
The important task in the design of a combined classifier is to select the suitable base models, which constitute its building blocks. The main contribution of this research work is an attempt to solve the problem of adaptively selecting the base learning vector quantisation (LVQ) models for recognition of human faces. A novel approach is presented for the automatic elimination of redundant hidden layer neurons of base models. Redundancy of a neuron is measured by the variance of the face image represented by that neuron of the neural network. Empirical comparison is performed between growing and pruning networks, i.e. networks generated through elimination of redundant or blind neurons. Results indicate that the elimination of blind neurons leads to base models which have less structural complexity and are more efficient than those generated using growing network.