Masked Face Recognition for Secure Authentication (2008.11104v1)

Abstract: With the recent world-wide COVID-19 pandemic, using face masks have become an important part of our lives. People are encouraged to cover their faces when in public area to avoid the spread of infection. The use of these face masks has raised a serious question on the accuracy of the facial recognition system used for tracking school/office attendance and to unlock phones. Many organizations use facial recognition as a means of authentication and have already developed the necessary datasets in-house to be able to deploy such a system. Unfortunately, masked faces make it difficult to be detected and recognized, thereby threatening to make the in-house datasets invalid and making such facial recognition systems inoperable. This paper addresses a methodology to use the current facial datasets by augmenting it with tools that enable masked faces to be recognized with low false-positive rates and high overall accuracy, without requiring the user dataset to be recreated by taking new pictures for authentication. We present an open-source tool, MaskTheFace to mask faces effectively creating a large dataset of masked faces. The dataset generated with this tool is then used towards training an effective facial recognition system with target accuracy for masked faces. We report an increase of 38% in the true positive rate for the Facenet system. We also test the accuracy of re-trained system on a custom real-world dataset MFR2 and report similar accuracy.

• The paper introduces MaskTheFace, an open-source tool that simulates masks on facial datasets to enable robust retraining of recognition systems.
• It retrains existing models like Facenet, achieving about a 38% increase in true positive rates for masked face recognition.
• Real-world evaluation on the MFR2 dataset confirms that the improved systems perform reliably for both masked and unmasked faces.

Masked Face Recognition for Secure Authentication

The proliferation of face masks due to the COVID-19 pandemic has posed significant challenges to existing facial recognition systems employed in various authentication applications. The paper "Masked Face Recognition for Secure Authentication" addresses an acute problem for systems that rely on facial recognition for tasks such as attendance tracking and device unlocking. These systems face reduced functionality when confronted with masked faces, a situation that threatens to invalidate pre-existing datasets and impair system operations.

Contributions and Methodology

The authors introduce a novel approach to alleviate this problem by augmenting current facial datasets, allowing recognition systems to identify masked faces without forming entirely new datasets. Key contributions include:

1. Development of MaskTheFace: An open-source tool designed to simulate masks on faces within existing datasets. It is capable of generating a wide variety of mask types, facilitating the creation of extensive masked face datasets.
2. Augmentation of Facial Recognition Training: The generated datasets are utilized to retrain existing systems, such as Facenet, to improve recognition accuracy for masked faces. The updated training process does not compromise the system's performance on unmasked faces.
3. Real-World Dataset Evaluation: The researchers curated a small dataset called MFR2, comprised of real-world masked faces for testing purposes. Results indicate that systems retrained with MaskTheFace-generated datasets perform comparably on real-world masked images.

Experimental Evaluation

The paper employs a robust experimental framework using both simulated and real masked datasets. The authors demonstrate the efficacy of their retraining approach using the following steps:

• Face Recognition System Selection: The Facenet architecture was chosen for evaluation, utilizing its ability to create face embeddings. The system was trained intensively on the VGGFace2-mini dataset with and without simulated masks.
• Dataset Description: They introduced VGGFace2-mini-SM, a dataset derived from VGGFace2-mini, enriched with simulated masks generated by MaskTheFace. Additionally, accurate performance was assessed on subsets of the well-known LFW dataset enhanced with simulated masks.
• Metrics for Evaluation: The facial recognition system's performance was quantified using metrics such as Max Accuracy, ACC @ FAR=0.1%, and TPR @ FAR=0.1%, providing a reliable assessment of system robustness under varying conditions.

Results

The numerical findings are significant. Retraining the Facenet system with MaskTheFace-generated datasets results in an approximate 38% increase in true positive rates for masked faces, indicating substantial performance restoration. Moreover, the results on the real-world MFR2 dataset reaffirm the system's capability, with modest performance differentials, thereby establishing the practicality of the proposed approach in real-world scenarios.

Implications and Future Directions

This research holds considerable implications for both the theoretical and practical domains of AI-based facial recognition. The use of MaskTheFace signifies a pivotal shift in accommodating unforeseen masking scenarios without overhauling existing datasets entirely—a method proving useful not only in current pandemic conditions but also as a foundational approach to managing occluded face scenarios.

Future research could explore further optimization of MaskTheFace, extending the range of mask types and face occlusion circumstances it can simulate. Additionally, investigating transfer learning techniques to optimize the computational aspects of retraining existing architectures can improve scalability. Sustainable development in this field will likely focus on adapting to new public health requirements while maintaining robust, user-friendly authentication systems.