Masked Face Recognition Dataset and Application (2003.09093v2)

Abstract: In order to effectively prevent the spread of COVID-19 virus, almost everyone wears a mask during coronavirus epidemic. This almost makes conventional facial recognition technology ineffective in many cases, such as community access control, face access control, facial attendance, facial security checks at train stations, etc. Therefore, it is very urgent to improve the recognition performance of the existing face recognition technology on the masked faces. Most current advanced face recognition approaches are designed based on deep learning, which depend on a large number of face samples. However, at present, there are no publicly available masked face recognition datasets. To this end, this work proposes three types of masked face datasets, including Masked Face Detection Dataset (MFDD), Real-world Masked Face Recognition Dataset (RMFRD) and Simulated Masked Face Recognition Dataset (SMFRD). Among them, to the best of our knowledge, RMFRD is currently theworld's largest real-world masked face dataset. These datasets are freely available to industry and academia, based on which various applications on masked faces can be developed. The multi-granularity masked face recognition model we developed achieves 95% accuracy, exceeding the results reported by the industry. Our datasets are available at: https://github.com/X-zhangyang/Real-World-Masked-Face-Dataset.

• The paper introduces a multi-granularity masked face recognition model that enhances accuracy from 50% to 95% by focusing on uncovered facial features.
• The paper presents three unique datasets: MFDD with 24,771 images, RMFRD with 5,000 masked and 90,000 unmasked images, and SMFRD with 500,000 simulated images.
• The paper outperforms existing technologies by improving accuracy compared to industry benchmarks, enabling robust and practical application in contactless verification systems.

Masked Face Recognition Dataset and Application

The paper "Masked Face Recognition Dataset and Application" addresses a crucial challenge that emerged prominently during the COVID-19 pandemic: the ineffectiveness of traditional facial recognition technologies when individuals are wearing masks. The authors from the National Engineering Research Center for Multimedia Software at Wuhan University propose a novel approach by developing and releasing a set of masked face datasets to advance research and application in this domain.

Key Contributions

This work introduces three distinct types of datasets to facilitate research in masked face recognition:

1. Masked Face Detection Dataset (MFDD): Sourced from existing research and internet-crawled images, this dataset contains 24,771 images, annotated to indicate mask presence. It serves as a resource for training models to detect masked faces.
2. Real-world Masked Face Recognition Dataset (RMFRD): Created using a python crawler fetching images of 525 individuals, RMFRD includes 5,000 masked and 90,000 unmasked images. It is considered the largest dataset of its kind, aiming to support the identification of individuals wearing masks.
3. Simulated Masked Face Recognition Dataset (SMFRD): This dataset derives from augmenting existing datasets like LFW and Webface by algorithmically overlaying masks on images. It encompasses 500,000 images from 10,000 subjects, broadening the diversity and robustness of training data.

Recognition Techniques

The authors developed a multi-granularity masked face recognition model which emphasizes uncovered facial features such as the eyes and eyebrows. By applying differential attention weights to visible face parts, the model significantly enhances recognition accuracy, reported to reach 95%. This improvement is substantial given the baseline recognition rate of 50% without such specialized training.

Comparative Analysis

Compared to existing solutions by technology enterprises such as Sense Time Technology and Hanvon Technology, which report accuracy rates of around 85%, the multi-granularity approach proposed in this paper presents a higher accuracy. The best industry-reported accuracy before this paper was 90% by MINIVISION Technology. Thus, the results in this paper advance the state of the art in masked face recognition.

Implications and Future Scope

The datasets provided offer a critical resource for further exploration and refinement of masked face recognition technologies. These datasets and methodologies enable practical applications in scenarios needing contactless verification, such as public transit security checks and workplace attendance systems. The persistent need for such technology, exacerbated by health concerns and environmental conditions like haze, underscores its long-term relevance.

The research sets a precedent for addressing masked face recognition’s practical challenges, particularly concerning occluded facial features. Future research can build upon these foundations, potentially integrating more sophisticated neural models or fusion techniques that could leverage additional biometric cues, such as gait or voice, to further enhance recognition performance under challenging conditions.

In conclusion, this paper contributes valuable datasets and a methodology that significantly improves masked face recognition accuracy, a pertinent topic given global health concerns, and paves the way for advancing both applied and theoretical research in the field.