Facial Expression Recognition Techniques and Comparative Analysis Using Classification Algorithms

Abstract

With the development of technology and hardware possibilities, it has become possible to analyze the changes that occur as a result of the reflection of emotional state on facial expression with computer vision applications. Facial expression analysis systems are used in applications such as security systems, early diagnosis of some diseases in the field of medicine, human-computer interaction, and safe driving. Facial expression analysis systems developed using image data consist of 3 basic stages. These are; extracting the face area from the input data, extracting the feature vectors of the data and classifying the feature vectors. In this study, a hybrid method for facial expression analysis is proposed. The method aims to combine the ability of deep learning models in feature extraction with the ability of machine learning to classify small datasets. Multi Task Cascaded Convolutional Network (MTCNN) has been used to detect the face region in the input data. The features extracted from the fully connected layer of the AlexNet model, which achieves successful results in classification problems, have been classified with K-Nearest Neighborhood (KNN), Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) algorithms. Machine learning and deep learning methods are widely used in facial expression analysis systems proposed in the literature. In this study, the performances of LDA, SVM and KNN algorithms have been analyzed using JAFFE dataset without data augmentation. With LDA, SVM and KNN algorithms, 89.2%, 88.3% and 87.8% accuracy has been achieved respectively.