Mask Attention-SRGAN for Mobile Sensing Networks.

Area of Science:

• Computer Science
• Biometrics
• Image Processing

Background:

• Biometric systems, including iris and face recognition, offer high accuracy for identity verification.
• Integrating low-cost image sensors into Internet of Things (IoT) devices reduces costs but often compromises the resolution needed for accurate biometric recognition.
• Maintaining high biometric accuracy with low-resolution sensors in mobile sensing networks is a significant challenge.

Purpose of the Study:

• To develop a single image super-resolution (SISR) algorithm capable of enhancing low-resolution biometric images.
• To improve the recognition accuracy of biometric systems utilizing low-cost image sensors.
• To address the trade-off between cost and performance in biometric identity recognition.

Main Methods:

• Proposed MA-SRGAN, a novel Generative Adversarial Network (GAN) based SISR algorithm incorporating a mask-attention mechanism.
• Modified the state-of-the-art nESRGAN+ model by adding a discriminator with an additional loss term.
• Focused the GAN's attention on the region of interest (ROI) for enhanced feature detail recovery.

Main Results:

• Experimental results on the CASIA-Thousand-v4 and Celeb Attribute datasets demonstrated the effectiveness of MA-SRGAN.
• The proposed method successfully recovered intricate details within crucial facial and iris regions.
• Enhanced image super-resolution led to significant improvements in biometric recognition accuracies.

Conclusions:

• MA-SRGAN effectively enhances low-resolution biometric images, improving recognition accuracy.
• The mask-attention mechanism in GANs is crucial for focusing on discriminative regions in biometric data.
• This approach offers a viable solution for maintaining high biometric performance in cost-sensitive IoT applications.