Deep facial expression detection using Viola-Jones algorithm, CNN-MLP and CNN-SVM

Computer vision researchers are now studying the process of recognizing emotions from facial expressions. Our system is based on his three-step method in this article, which includes face detection, feature extraction, and classification. Capture a photo/video to get facial recognition information and find the face area in this image. Face extraction uses the Viola-Jones algorithm to find reflective areas (eyes, mouth, nose, and temples) in specific faces. In order to extract the faces, we have built a database of frontal face images. We offer two systems. The first facial emotion detection system is based on classification using raw facial images, and the second extracts the oriented gradient histogram (HOG) from facial images. For the classification phase, we use three classifiers: support vector machines (SVM), Convolutional Neural Network (CNN) and hybrid CNN-SVM. To increase the performance of our facial emotion recognition system, we propose to merge the two CNN outputs of the two systems to create deep features that are merged as inputs of two classifiers (MLP and SVM). The experiments are performed the Ryerson Multimedia Laboratory (RML) dataset. The objective is to compare the performances of these methods and to identify the most suitable approach. Our experimental results showed good accuracy compared to previous studies.