Robust Face Recognition Using the Deep C2D-CNN Model Based on Decision-Level Fusion.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Face recognition relies on multiple features for robust identification.
• Traditional and deep learning methods struggle with significant differences between test and training datasets.
• Feature fusion is crucial for enhancing face recognition performance.

Purpose of the Study:

• To propose a novel method, C2D-CNN, for improved face recognition.
• To enhance CNN architecture for faster convergence and better feature preservation.
• To address low recognition accuracy caused by dataset discrepancies.

Main Methods:

• Developed C2D-CNN, integrating color 2-dimensional principal component analysis (2DPCA) with convolutional neural network (CNN) features.
• Implemented a new CNN model featuring a normalization layer, layered activation function, and probabilistic max-pooling.
• Employed decision-level fusion of learned features.

Main Results:

• C2D-CNN demonstrated superior performance compared to state-of-the-art methods.
• The proposed CNN enhancements accelerated network convergence and training time.
• The method effectively improved recognition accuracy, mitigating dataset differences.

Conclusions:

• C2D-CNN offers a significant advancement in face recognition technology.
• The novel CNN architecture contributes to more efficient and accurate models.
• This approach provides a robust solution for face recognition challenges posed by dataset variations.