Identificação biométrica ocular através de redes neurais siamesas

Abstract

The use of technology to make processes more efficient has led to a growing need for automatic identification of individuals. Traditional identification methods, such as signatures and cards, are becoming less effective and are being replaced by biometric identification methods, which use physical or behavioral characteristics of individuals. This work proposes the development of iris-based biometric recognition models, which use machine learning techniques, specifically Siamese and convolutional networks. Two Siamese network models were built, one simpler and the other more robust. Experiments performed on a database containing iris images under various capture conditions, such as noise and occlusions, provided a model with an accuracy of 84.51% on test data.