Many reports on the use of a multi-layer perceptron (MLP) based neural network to classify biometric data exist in the literature. The technique reported on in this paper, however, is not limited to the classifier task; it also includes the preprocessing and post-processing steps, which makes it particularly valuable as both a teaching vehicle for classifiers and as an example of a system-level implementation.
The biometric features selected are unusual—hand geometry “parameters”—and the authors make a good case of demonstrating their suitability. Two techniques employed are novel: the normalization of the input data for the MLP classifier, and the post-processing, or rather the mixture of post-processing and MLP result representation, using block error correcting codes (of the same family as Reed-Solomon codes).
Details on the data acquisition procedure and the final decision presentation complete this carefully crafted report. This report should be of value to biometric systems designers, and to those studying the application of pattern recognition methods.