![](bg7.png)
Also to improve the accuracy even higher, We
have a few proposed ideas such as mask saturation
randomization to eliminate bias for different illumi-
nated environments or to expand the dataset. Not
to mention, one of the state-of-the-art CNN mod-
els might yield completely different results and tests
should be done to be determined.
5 CONCLUSION
Ultimately, in this paper, we had proposed a CNN
artificial intelligence solution for automatic analysis
of a person’s facial expression and determination.
Which had seven different expression categories that
are ’Angry’, ’Disgust’, ’Fear’, ’Happy’, ’Neutral’,
’Sad’ and lastly ’Surprise’.
One reason so many expression categories have
been used is to have a paper that is not too similar to
others. Even though some expressions are incredibly
hard to make out with a face mask even from a hu-
man’s perspective. Although, reducing the number of
emotions would have a dramatic increase in test accu-
racy, we would not want to downgrade our goal. To
accomplish the objectives in this paper we have used
Python in tandem with Google Colab and these two
combined to offer an incredible workspace for devel-
opers.
The paper consisted of three main parts, the syn-
thetic dataset generation since there is a lack of
masked face datasets as of writing this. The CNN
model was built and trained with our synthetically
made dataset that we generated by using a pre-
existing expression dataset with no masks. Subse-
quently, generating and saving the trained model’s
weights and multipliers. Ultimately, to be used in a
video feed to offer real-time FER with face masks.
6 FUTURE WORK
We do believe that this work still has more space to
grow. More features could be added to improve al-
ready existing systems. One other system that would
be beneficial to add for example would be saturation
randomization for the face masks in the dataset gener-
ator. Since the dataset consists of only black or white
masks, the model might have a slight bias towards
those two colors. However with a saturation random-
ization system it would be equivalent to color ran-
domization since the images are processed in a single
gray-scale layer, a saturation change would be com-
parable to a hue change in a full-color image.
REFERENCES
Ari, N. and Ustazhanov, M. (2014). Matplotlib in python.
In 2014 11th International Conference on Electronics,
Computer and Computation (ICECCO). IEEE.
Bisong, E. (2019). Building machine learning and deep
learning models on Google cloud platform. Springer.
Brownlee, J. (2016). Deep learning with Python: develop
deep learning models on Theano and TensorFlow us-
ing Keras. Machine Learning Mastery.
Castellano, G., De Carolis, B., and Macchiarulo, N. (2022).
Automatic facial emotion recognition at the covid-19
pandemic time.
Christ, M., Braun, N., Neuffer, J., and Kempa-Liehr, A. W.
(2018). Time series feature extraction on basis of scal-
able hypothesis tests (tsfresh–a python package). Neu-
rocomputing.
Clark, A. et al. (2015). Pillow (pil fork) documentation.
readthedocs.
Das, A., Wasif Ansari, M., and Basak, R. (2020). Covid-
19 face mask detection using tensorflow, keras and
opencv. In 2020 IEEE 17th India Council Interna-
tional Conference (INDICON).
DeTone, D., Malisiewicz, T., and Rabinovich, A.
(2016). Deep image homography estimation. CoRR,
abs/1606.03798.
Henke, L., Guseva, M., Wagemans, K., Pischedda, D.,
Haynes, J.-D., Jahn, G., and Anders, S. (2022). Sur-
gical face masks do not impair the decoding of facial
expressions of negative affect more severely in older
than in younger adults.
Howse, J. (2013). OpenCV computer vision with python.
Packt Publishing Birmingham.
King, D. E. (2009). Dlib-ml: A machine learning toolkit.
The Journal of Machine Learning Research.
Kobayashi, H. and Hara, F. (1992). Recognition of six basic
facial expression and their strength by neural network.
In [1992] Proceedings IEEE International Workshop
on Robot and Human Communication. IEEE.
Sharifara, A., Mohd Rahim, M. S., and Anisi, Y. (2014).
A general review of human face detection including a
study of neural networks and haar feature-based cas-
cade classifier in face detection. In 2014 International
Symposium on Biometrics and Security Technologies
(ISBAST).
Singhal, A. (2021). Facial Expression Dataset. Kaggle.
Van Der Walt, S., Colbert, S. C., and Varoquaux, G. (2011).
The numpy array: a structure for efficient numerical
computation. Computing in science & engineering.
Viola, P. and Jones, M. J. (2004). Robust real-time face
detection. International Journal of Computer Vision.
Wu, B.-F., Chen, B.-R., and Hsu, C.-F. (2021). Design of
a facial landmark detection system using a dynamic
optical flow approach. IEEE Access.
Yang, B., Jianming, W., and Hattori, G. (2021). Face mask
aware robust facial expression recognition during the
covid-19 pandemic. In 2021 IEEE International Con-
ference on Image Processing (ICIP).
Yang, B., Wu, J., and Hattori, G. (2020). Facial expression
recognition with the advent of face masks. MUM ’20.
Facial Expression Recognition with Quarantine Face Masks Using a Synthetic Dataset Generator
205