Face and Body-Based Human Recognition by GAN-Based Blur Restoration

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Long-distance human recognition in indoor settings often combines face and body information due to camera placement limitations.
• Motion blur significantly degrades face image quality in close-range indoor environments, with limited research on deblurring body images.

Purpose of the Study:

• To address motion blur challenges in indoor face and body recognition.
• To propose a novel method for restoring blurred images and fusing features for improved recognition accuracy.

Main Methods:

• Utilized a generative adversarial network (GAN) to restore blur in both face and body images.
• Employed a deep convolutional neural network (CNN) to extract features from restored images.
• Fused matching scores derived from face and body features for recognition.

Main Results:

• The proposed method achieved an equal error rate (EER) of 7.694% on the Dongguk face and body dataset version 2 (DFB-DB2).
• The method obtained an EER of 5.069% on the ChokePoint dataset.
• Demonstrated superior performance compared to existing state-of-the-art methods.

Conclusions:

• The GAN-based image restoration and CNN-based feature fusion effectively mitigates motion blur issues in indoor human recognition.
• The proposed approach significantly enhances recognition accuracy, outperforming current methods on benchmark datasets.