Enhanced Facial Expression Recognition Using Pre-trained Models and Image Processing Techniques

Facial expressions are a type of nonverbal communication and one of the most important activities in effective human-computer interaction. In recent years, facial expression recognition has been a hot topic for researchers particularly for smart applications. However, Facial Expression Recognition (FER) still faces many challenges, such as non-uniform illumination, aging, pose variations, etc. The proposed methodology is based on deep learning technique for extracting and classifying the features from images. Four datasets are utilized to evaluate the effectiveness of the suggested methodology which are CK+, JAFFE, RAFDB, and FER2013. The facial part is detached, cropped, and a CLAHE filter is applied to improve contrast in images. We use pre-trained models of VGG16, Inception V3, MobileNet, and DenseNet but replace their fully connected layers with our own suitable ones. The accuracy achieved on FER3013, RAFDB, CK+, and JAFFE datasets were 90.72%, 93.12%, 97.50%, and 93.55% respectively.