Enhanced multimodal biometric recognition systems based on deep learning and traditional methods in smart...

Area of Science:

• Biometric Security
• Data Security
• Pattern Recognition

Background:

• Biometric security is crucial for data security, with multimodal systems offering enhanced accuracy.
• Integrating electrocardiogram (ECG) signals and fingerprint data presents an effective multimodal approach.
• Challenges remain in developing accurate and reliable multimodal biometrics for smart environments.

Purpose of the Study:

• To evaluate unimodal and multimodal biometric systems using Convolutional Neural Networks (CNN).
• To compare traditional methods with deep learning approaches for fingerprint and ECG fusion.
• To assess the effectiveness of parallel and sequential fusion strategies in multimodal biometrics.

Main Methods:

• Developed and compared unimodal and multimodal biometric systems using CNN.
• Employed various feature extraction and classification techniques for fingerprint and ECG fusion.
• Evaluated systems on MIT-BIH (ECG) and FVC2004 (fingerprint) databases, including virtual datasets with and without augmentation.

Main Results:

• Sequential multimodal biometrics achieved an optimal Area Under the ROC Curve (AUC) of 0.99.
• Parallel multimodal biometrics achieved an AUC of 0.96, outperforming unimodal fingerprint biometrics (0.87 AUC).
• The proposed CNN models, especially for ECG and sequential fusion, demonstrated superior performance compared to other systems.

Conclusions:

• Multimodal biometrics, particularly using CNN, significantly outperform unimodal systems in accuracy and detection rates.
• Sequential fusion of ECG and fingerprint data shows exceptional promise for robust biometric security.
• The proposed CNN-based approach offers a powerful solution for advanced biometric authentication in smart environments.