Abstract:
Face expression is an important feature for a human’s mental and initial state of
face-to-face conversation in daily life. In recent years, artificial intelligence has become
popular in computer vision and image processing. Facial expression classification is a
critical field of study in computer vision and artificial intelligence, with applications
from human-computer interaction, customer satisfaction, healthcare, video
conferencing, face animation, and behavioral psychology.
This proposed system works in three parts, including preprocessing, feature
extraction, and classifier of the image. Firstly, the image preprocessing includes
resizing, and normalization. Then, Local Binary Pattern (LBP) and convolutional layer
of Visual Geometry Group (VGG) 16 network are used to extract features. For the better
classification result, the extracted feature vectors are combined and are applied as input
to the convolutional neural network model to classify for facial expression
classification. Finally, the model is used to classify five emotions such as happiness,
sadness, anger, disgust, and surprise. For the improvement of performance model, the
model has used Adam optimizer, Batch normalizing, and Softmax activation functions
to prevent overfitting the datasets during training. The experimental result of the
proposed model was tested on Extended Cohn Kanade (CK) dataset and obtained a 99%
accuracy rate.