Robust face recognition based on multi-task convolutional neural network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Face recognition accuracy is challenged by variations in lighting, complex backgrounds, and lack of prior image data.
• Existing methods may struggle with diverse and imperfect facial image datasets.

Purpose of the Study:

• To develop and evaluate a novel face detection and recognition algorithm using a multi-task convolutional neural network (MTCNN).
• To address the limitations of current face recognition systems in real-world scenarios.

Main Methods:

• The proposed MTCNN utilizes three cascaded networks with a candidate box and classifier approach for efficient face recognition.
• Performance was evaluated against Region-CNN (R-CNN) and Faster R-CNN using a custom 50-face database.
• Metrics included Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measurement (SSIM), and Receiver Operating Characteristic (ROC) curves.

Main Results:

• MTCNN demonstrated superior performance with higher average PSNR (1.24 dB over R-CNN, 0.94 dB over Faster R-CNN) and SSIM (10.3% over R-CNN, 8.7% over Faster R-CNN).
• MTCNN achieved a significantly higher Area Under the Curve (AUC) of 97.56% compared to R-CNN (91.24%) and Faster R-CNN (92.01%).
• The algorithm proved effective even for face images with defective features.

Conclusions:

• The MTCNN algorithm offers a substantial improvement in face recognition accuracy and a reduction in false detection rates.
• This enhanced face recognition capability can bolster security in critical areas and improve operational efficiency by reducing manual resource allocation.