M2FRED

Mobile Masked Face REcognition through periocular Dynamics analysis dataset

The outbreak of novel coronavirus 2019 (COVID-19) has rapidly required the development of brand-new biometric recognition technologies able at analyzing subjects which wear sanitary and surgical masks, especially in high security applications. Indeed, in this challenging context, the problem of face recognition is equivalent to the periocular recognition combining several biometrics traits such as iris, upper and lower eyelids, eye folds, and eye corners, skin texture, fine wrinkles, color, skin pores, sclera etc. Face dynamics, i.e. the morphing of facial surfaces due to facial muscles, must still be considered since they can significantly alter the face appearance and so introducing error in biometric recognition approaches which are based on static traits. Notwithstanding the enormous potential of the ocular traits for controlled and uncontrolled scenario, actually their possible fusion still needs to be more investigated to develop a masked face recognition technique in different scenery such us mobile, near infrared (NIR) and thermal camera, surveillance cameras. Also, the presence of a mask occluding part of the face allows to focus on those side regions which can provide acceptable degrees of reliability, at least in verification mode. Such technologies will reveal their accessibility also in daily professional scenarios, like hospitals, where the sanitary masks are worn whether or not the coronavirus (by surgeons or operators in infectious disease departments), or building sites (in order to avoid the inhalation of dust) as well as in more practical contexts to verify the accesses to restricted/controlled areas like airports, banks, postal offices and similar. 

The BIPLab has then made public the dataset M2FRED containing 43 subjects with the following data structure:

  • subjects ID number from 000 to 043
  • a folder for each subject: one without mask (i.e. 000_0) one with mask (i.e. 000_1)
  • 16 videos for each folder: 8 indoor/8 outdoor

NB: Recommended player VLC Media player.

Available for download here

The dataset is password-protected. To access its content, please send an email to biplab@unisa.it