Multimodální rozpoznání probandů vybraných variabilních markantů s využitím metod AI

Abstract

This bachelor thesis deals with the issue of deep learning and artificial intelligence in the context of multimodal biometric systems with a focus on the GoogLeNet architecture. The introductory part defines the theoretical framework of anatomical biometric features, principles of convolutional neural networks, and the importance of hyperparameters for the learning process. The purpose of the thesis is an experimental analysis of the influence of selected hyperparameter configurations on the stability and accuracy of face, palm, and fingerprint identification. The second half of the work is dedicated to fulfilling this goal, where the results of individual scenarios are classified, and the possibilities of future development are discussed in the conclusion.