New Transformer Model Enhances Face Recognition with Masked Faces

Researchers have developed PLGSA-Transformer, a novel framework for face recognition that addresses the challenges posed by facial masks. This system utilizes periocular landmark-guided spatial attention to focus on visible facial regions around the eyes and forehead, integrating features from EfficientNetB3. A hybrid CNN-Transformer architecture processes these features, and an occlusion-adaptive cosine threshold adjusts matching scores based on predicted occlusion severity. The model demonstrated high accuracy, achieving 97.22% pair verification accuracy on a dataset comprising masked and unmasked faces, outperforming previous methods. AI

IMPACT This research offers a more robust solution for face recognition in real-world scenarios where masks are common, potentially improving security and identification systems.

RANK_REASON Academic paper detailing a new model and its performance on a specific task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Abdullah
• Adnan
• COVID-19
• EfficientNetB3
• Kaggle CelebA-HQ masked collection
• MediaPipe
• PLGSA-Transformer
• Shnain
• VGG-16
• Zenodo MDMFR

• paper
• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →