Face Expression Recognition with Local Ternary Pattern Images using Convolutional Neural Network and Extreme Learning Machine

Facial Expression Recognition (FER) performed computationally is an exciting task to explore in computer vision. Some methods have been proposed to handle variants of illumination in FER. Based on research, Local Ternary Pattern (LTP) as a feature extractor can handle the variation of lights. However, LTP is a traditional feature extractor that needs to be processed manually. Unlike LTP, Convolution Neural Network (CNN) architecture has an automatic feature extractor. Therefore, this study proposes LTP images as input into CNN architecture to handle light variations and keep the feature extraction automatically. Afterward, in the classification layer, Extreme Learning Machine (ELM) is employed as a classifier to improve the training speed of the original CNN classifier. The proposed model performance for the KDEF dataset with 10-fold cross-validation yields an accuracy of 85.51%