Enhancing The Performance Of Neural Network Classifiers Using Selected Biometric Features
Heman Mohabeer, K. M. S. Soyjaudah, Narainsamy Pavaday · International Conference on Sensor Technologies and Applications · 2011
This paper describes an application which increases the overall efficiency of a neural network classifier intended for authentication whilst using fewer biometric features. Normalization of the biometric data is generally performed to remove unwanted impurities. However, in this case, when performing normalization, the statistical property for each set of data has also been taken into consideration prior to the classification process. Combination of the normalized biometric features has been performed while comparing their standard deviation. The resulting fused data has correlation value as low as possible. This gives rise to a higher probability of uniquely identifying a person in feature space. The proposed system is intended to make authentication faster by reducing the number of biometric features without degrading the overall performance of the classifier. The performance of the classifier was computed using the mean square error (MSE). The results show that redundant biometric data can indeed be excluded without degrading the performance of the classifier.