.png){kind=icon}
.png){kind=icon}
Return
A multi-classifier system for automatic fingerprint classification using transfer learning and majority voting
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-15337-6.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Fingerprints fulfill a critical role in the context of community safety and criminal investigations, especially for forensic investigations, law enforcement, border access and security. For fingerprint analysis, a variety of machine learning and neural network techniques have been presented. Various AI-driven approaches are available to conduct automated activities to maintain biometric systems at the forefront of technological development. In this paper, we propose a framework for automatic classification of fingerprints that combines deep transfer learning and a majority voting system. Our multi-classifier system is capable of efficiently classifying six different types of fingerprints. We build 16 commonly used deep transfer learning models by training them with fingerprint datasets, resulting in 16 fingerprint classifiers. The training and testing phases involve six distinct fingerprint databases consisting of both real and synthetic images. We created sets of fingerprint classifiers and applied a soft majority voting method to each set, aiming to identify the optimal combination of classifiers. This allowed us to determine the most effective set of classifiers for fingerprint classification. The results of our experiments indicate that the majority voting approach of three deep transfer-learning models, namely DenseNet 121, ResNet152V1, and EfficientNetB7, outperforms the individual transfer learning structures in terms of precision in fingerprint classification.
Keywords:
Automatic classification
Fingerprint classification
Mlti-Classifier system
Transfer learning
Majority voting
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W