Facial Expression Recognition using Multi-Feature Concatenation of Local Face Components and Hierarchical SVM
Mahdi Jampour, Amin Karimi Sardar · 2021
To prepare an efficient and reliable automatic Facial Expression Recognition (FER) system, 1) face and active regions detection, 2) feature extraction, and 3) feature classification have to be robustly employed. In this paper, we propose a novel approach with three contributions. First, we have used facial local active regions such as Mouth, Eyes, and Eyebrows because they are more affected by emotions. Second, very efficient joint feature descriptors, HOG (Histogram of Oriented Gradients) and LDP (Local Directional Pattern) are employed to extract features. Finally, a hierarchical multi-class classification SVMs model has been designed for classification. The results show that our approach can efficiently recognize human facial expressions from facial images where it can outperform the related work. Moreover, we show that our approach is stable on illumination variation, which could be useful for evaluating the facial with variant skin tone or facial images in different intensity conditions.