研究目的
Investigating the effectiveness of a novel deep CNN framework for biometric verification using physiological face and iris traits.
研究成果
The proposed unimodal biometric verification system performed well, serving tighter security needs, and the multimodal biometric system achieved 100% accuracy. The study highlights the superiority of deep CNN models over conventional methods in feature extraction and classification, suggesting promising directions for future research in non-deterministic problems.
研究不足
The study acknowledges the challenge in generalizing the deep learning CNN parameters such as selection of optimal output filters, number of epochs, batch size, and learning rate for physiological and behavioral biometric modalities.
1：Experimental Design and Method Selection:
The study proposed a novel deep CNN architecture for feature extraction in two convolution layers, focusing on biometric verification using face and iris traits.
2：Sample Selection and Data Sources:
Utilized ORL dataset for face and CASIA dataset for iris, considering 40 users.
3：List of Experimental Equipment and Materials:
Not explicitly mentioned.
4：Experimental Procedures and Operational Workflow:
The CNN architecture included convolution, ReLU, and max pooling layers, with specific parameters like batch size, epochs, and learning rate adjusted for optimal performance.
5：Data Analysis Methods:
The performance was evaluated based on Genuine Acceptance Rate (GAR) at standard False Acceptance Rate (FAR) thresholds.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
