Grip-Pattern Recognition in Smart Gun Based on Likelihood-Ratio Classifier and Support Vector Machine

In the biometric verification system of a smart gun, the rightful user of a gun is recognized based on grip-pattern recognition. It was found that the verification performance of this system degrades strongly when the data for training and testing have been recorded in different sessions with a time lapse. This is due to the variations between the probe image and the gallery image of a subject. In this work the grip-pattern verification has been implemented based on both classifiers of the likelihood-ratio classifier and the support vector machine. It has been shown that the support vector machine gives much better results than the likelihood-ratio classifier if there are considerable variations between data for training and testing. However, once the variations are reduced by certain techniques and thus the data are better modelled during the training process, the support vector machine tends to lose its superiority.