A deep feature-level fusion model for masked face identity recommendation system.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Airborne outbreaks and pollution necessitate mask-wearing, increasing challenges for facial recognition systems.
• Existing facial recognition models struggle with mask-obscured faces, limiting their effectiveness in real-world scenarios.
• Criminals exploit masks to conceal identity, posing security risks.

Purpose of the Study:

• To develop a robust deep-learning framework for accurate facial identity recognition of individuals wearing masks.
• To address the limitations of current facial recognition systems in handling mask-obstructed faces.
• To enhance security applications by enabling suspect identification from surveillance footage.

Main Methods:

• Proposed a deep-learning-based feature-fusion framework named FIREC.
• Integrated demographic-estimated features (age, gender, race) to compensate for lost facial information.
• Developed a system to provide a ranked list of potential identities for masked faces.

Main Results:

• Achieved high accuracy in recognizing both bare (99.34% Hit@10) and masked faces (97.65% Hit@10).
• Demonstrated the framework's effectiveness in compensating for information loss due to mask obstruction.
• Validated the system's capability to recommend potential identities from masked facial images.

Conclusions:

• The FIREC framework offers a significant advancement in masked face recognition technology.
• The system is highly beneficial for high-recall facial identity recognition applications, particularly in identifying suspects from CCTV footage.
• FIREC provides a viable solution for security challenges during public health crises when mask usage is prevalent.