e-mail: mc9 at cec dot wustl dot edu
|
|
Mei Chen
e-mail: mc9 at cec dot wustl dot edu |
|
Thresholding Method for Reduction of Dimensionality: Biometric Application Often recognition systems must be designed with relatively a small amount of training data. Plug-in test statistics suffer from large estimation errors, often causing the performance to degrade as the measurement vector dimension increases. To cope with this problem, we introduce the thresohlding method for dimensionality reduction in recognition systems. In the evaluation of Laser Doppler Vibrometry (LDV) as a candidate of a novel biometric marker, we first apply a time-frequency decomposition analysis on the training data, and each model of an individual for recognition is represented in a reduced dimensionality obtained by applying a threshold function on the information rate of each time-frequency component. The measure of information rate is based on the relative entropy. We obtain encouraging results based upon this approach. The implementation of relative entropy and dimensionality reduction method is discussed. |
|