An Investigation on Facial Emotional Expression Recognition Based on Linear-Decision-Boundaries Classifiers Using Convolutional Neural Network for Feature Extraction

This paper presents an investigation study on facial emotional expression recognition. Three linear-decision-boundaries classifiers: linear support vector classification (LSVC), linear discriminant analysis (LDA) and softmax (SM) techniques are utilized based on convolutional neural network (CNN) for efficient feature extraction, namely CNN-LSVC, CNN-LDA and CNN-SM respectively. Hyper-parameter tuning or selection needs the least effort for such three linear-decision-boundaries classifiers. In order to enhance recognition performance, particular image preprocessing: intensity transformation as well as image cropping technique are implemented before feeding input images into CNN feature extraction. Relying on 10-fold cross validation of 80%-20% training-testing CK+ dataset, above 90% average results of precision, recall, F1 scores and accuracy rates are yielded by all such three investigated methods. Confusion matrix is also determined for more-detail of results examination.