Implementation of Facial Expression Recognition (FER) Using Convolutional Neural Network (CNN)

Humans use facial expressions as a non-verbal medium of communication which commonly reflects how they are doing and their mood. The present study benefits from facial expressions by having a Facial Expression Recognition (FER) system that recognizes such expressions and produces an output matching each expression. A Convolutional Neural Network (CNN) model is trained to perform the facial expression recognition task. This model would distinguish six (6) types of facial expressions including anger, happy, sadness, fear, surprised, and neutral. Firstly, the trained CNN model was imported into an integrated development environment and a webcam was employed to record the user's facial expression in real-time. Next, the model will recognize the video recording captured by the webcam once it is fed into the system. With each recognized expression, a counter would be incremented. As the counter approached the minimal criteria, an output in the form of a recommendation would be displayed to the user. Based on the testing conducted, the developed FER system achieved an average accuracy of 93.61%. In addition, the specificity and the sensitivity scores obtained are 95.93% and 80.00%, respectively. This indicates the implementation of CNN in recognizing facial expressions is promising and convincing. In the future, it is preferable to use a dataset with a more significant and equal number of images in various circumstances.