Fingerprint Authentication Using Modular Neural Nets

Vitaly KrivenkoVitaly Krivenko graduated from Kharkiv National University of Radio Electronics, Ukraine and has successfully passed his Master’s viva at the Wessex Institute of Technology with his thesis entitled “Fingerprint authentication using modular neural nets”. The external examiner was Dr Rinat Khoussainov and the internal examiner was Prof Viktor Popov.

Vitaly performed research on neural networks and their ability in complex pattern recognition. The novelty of his research was fingerprint recognition using neural networks. The advantages of this approach are - no need for complex algorithms and simplification of the fingerprint recognition process.

During his research different neural networks and their configurations were evaluated specially for fingerprint recognition. A brief overview of different biometrics and the traditional fingerprint recognition methods was performed. The neocognitron was found eligible for fingerprint recognition as it is capable of recognition of rotated, shifted or scaled patterns.

To perform fingerprint recognition Vitaly firstly determined the correlation between the neocognitron parameters and its ability in complex pattern recognition. Several different shapes exhibiting structure similar to a fingerprint were taken as the ones to train and test the neocognitron. The results were summarized and the corresponding parameters were adjusted in order to elicit a greater network response for a fingerprint shape.

The thesis contained many different graphs showing statistical analysis of different neocognitron configurations performance on database containing 1000 fingerprints. Analysis of these graphs revealed merits and demerits of fingerprint recognition by means of the neocognitron.

At the end of his thesis a performance evaluation was carried out of the traditional approaches for fingerprint recognition and the fingerprint recognition performed by neocognitron.

The results obtained showed that the neocognitron was capable of fingerprint recognition with a quite low FAR value. In terms of fingerprint quality (quantity of misclassified images) the traditional approaches performed a little better than the tested neocognitron network configurations. But in terms of recognition speed the neocognitron performed better than traditional fingerprint recognition approaches. The results were obtained for non parallelized algorithms and parallelizing neocognitron is believed to speed up recognition process.

The research revealed the neural networks abilities in complex pattern recognition. This research may be of interest  to other researchers who study neural networks abilities and applications.

As a result of his research, both examiners recommended that Vitaly be awarded the degree of Master of Philosophy.

The support provided by Foreign and Commonwealth Office in the UK is gratefully acknowledged.