A system for identifying a person using biometrics technology and artificial neural network systems implemented using software simulation has been developed. The system extracts the features of the major creases in a person’s palm from its digitized image and verifies the identity of the person against a list of enrolled identities. The software simulation was implemented using hierarchical neural network architecture consists of a self organizing map (SOM) to select and extract features of the palm creases from the digitized image and an association layer as feature map classifier. The SOM makes use of competitive learning algorithm while the association layer uses back propagation neural network architecture through supervised learning. A unique set of codes was been generated out of the features of the palm creases of each sample individual. With the neural network’s inherent capability for pattern recognition, the system was able to generate similar codes, if not exactly the same codes for palm images belonging to the same individual. This became the basis for identity verification of the system.
Opportunities for further research may be made on several techniques that were discovered during the course of this study. Heuristic noise reduction techniques were applied for image processing and enhancement. Successful masking techniques were implemented to further enhance the image by eliminating the remaining ridges. The success of experimentally applying buffer zones in post processing routine eliminated false rejection and false acceptance errors. Modifications to the SOM algorithm proved to be effective in this particular application where normalization of the weight vectors based on distance-computed weight of neighboring nodes was applied. The seeds for these research opportunities have been established in this research and may be developed for other applications.
Opportunities for further research may be made on several techniques that were discovered during the course of this study. Heuristic noise reduction techniques were applied for image processing and enhancement. Successful masking techniques were implemented to further enhance the image by eliminating the remaining ridges. The success of experimentally applying buffer zones in post processing routine eliminated false rejection and false acceptance errors. Modifications to the SOM algorithm proved to be effective in this particular application where normalization of the weight vectors based on distance-computed weight of neighboring nodes was applied. The seeds for these research opportunities have been established in this research and may be developed for other applications.
