Robust Facial Expression Classification using CNN and Multi-Layer Perceptron Network classifiers

Abstract

Facial expression is the display of people's emotions, and it plays an important part in everyday conversation. As a result, Facial Expression Recognition (FER) is becoming an extremely important activity in modern society. Face identification, image extraction, and classification are the three phases of FER. While many FER systems are proposed to recognize only some face expressions, this paper proposes a recognition system for automatic face expression which recognizes all eight essential facial expressions (Normal, Cheerful, Anger, Scorn, Surprise, Sad, Terror, and Disgust). The approach is tested with the Extended CK+ dataset. The algorithm Viola-Jones is used for facial recognition in the presented process. A descriptor used for characteristics of descriptive face pictures is the Histogram of Oriented Gradients (HOG). The Primary Component Analysis (PCA) is used to achieve the most critical features to minimize feature dimensions. Finally, using two different classifiers, Multi-Layer Perceptron Neural Network (MLPNN) and Convolutional Neural Network (CNN), the presented approach categorized facial expressions, and the results were compared.