Table 1.
Advancements and complications of classical masked face recognition techniques.
| Author [citation] | Techniques | Advancements | Difficulties |
|---|---|---|---|
| Sumathy et al.18 | Double Generator Network | It accurately recognizes the face behind the mask by splitting the input images into higher and lower frequency attributes | This approach suffers from shift invariance problems |
| Kocacinar et al.19 | CNN |
It does not require plenty of samples for performing masked face recognition It identifies the incorrect usage of masks in the face |
It does not rectify the Interpretability challenges and Computational complexity issues |
| Zhang et al.20 | AMaskNet |
It utilizes a contribution estimator and feature extractor to detect masked faces at low cost It performs simple matrix multiplications to refine the feature representation of a face |
The generalization capacity of the model is low and gradient issues are high |
| Ullah et al.21 | DeepMaskNet |
It prevents the overfitting issues during the classification phase by using a dropout technique It retrieves deep, discriminative, and descriptive features for recognizing the masked faces |
Labeled data requirement is high |
| Faruque et al.22 | CNN | It employed an operational efficiency depth-wise normalization and batch normalization to progress the operational efficiency of masked face detection |
The performance of the framework decreases with respect to the increase in data The superiority of the model is low when there is variation or changes in face angles |
| Golwalkar and Mehendale23 | FaceMaskNet-21 | It has the ability to detect masked faces in static video files, live video streams, and static images |
It does not handle occlusions in images This model struggles to detect the masked face when the face angle is different in the input frame |
| Vu et al.24 | CNN |
It has high feasibility in retrieving facial features It retrieves only the relevant features to obtain accurate recognition outcomes |
The energy consumption of this network is high due to the usage of more parameters |
| Eman et al.25 | RPCA and MobileNetV2 |
It analyzes the key facial features to find the location and presence of masks on a face It solves the occlusion issues |
Accuracy is low while detecting faces in images with bad lighting conditions Different kinds of occlusions like hats and scarves cannot be removed in this model |