# Efficent masked face recognition method during the covid-19 pandemic 論文筆記
[論文連結](https://www.researchgate.net/publication/347874831_Efficient_Masked_Face_Recognition_Method_during_the_COVID-19_Pandemic)
## Abstract
Propose a method based on occlusion removel and deep learing-based features in oreder to address the problem of the masked face recognition process.
* The first step is to **remove the masked face region**.
* Next, apply 3 pre-trained deep CNN (VGG-16, AlexNet, ResNet-50), and use them to extract deep features from the obtained regions (**mostly eyes and forehead regions**).
* Finally, apply **MLP** for the classification prcess.
* **Has high recognition perfomance on RWMFD**(real world masked face dataset).
## Introduction
wearing tha mask face causes the following problems:
* exist face recognition mehods are not efficient when wearing a mask which cannot provide the whole face image for description.
* exposing the nose region is very a important part in face recognition.

Thus, they distinguish two different tasks, ***face mask recongion*** and ***masked face recognion***.
Face mask recongion is to check whether the person is wearing a mask.
Masked face recognion iaims to recognize a face with a mask based on **the eyes and the forehead regions**.
## Related works
## Motivation and contribution of the paper
## The proposed method
It has four steps:
### (1) Preprocessing and cropping filter
The dataset are already cropped around the face, so they **don't need a face detection**.(**it means that we should do it**)
But they need to correct the rotation. Using [Dlib-ml](https://www.jmlr.org/papers/volume10/king09a/king09a.pdf) to do it.

The next step is to apply a cropping filter in order **to extract only the non-masked region**.
(1) normalize all faces into 240 * 240pixels
(2) partiton a face into 100 blocks (a block is 24*24 picxels)
(3) extract number 1-50 (like picture 3 below)

### (2) Feature extraction layer
To extract deep features from informative regions, they have employed 3 pre-trained models as featue extractors:
VGG-16
Alex-Net
ResNet-50

### (3) Deep bag of features layer
### (4) Fully connected layer and classification