CNN-SVM Fusion: A Robust Method for Emotion Detection in Faces

Analysis of facial expressions for emotion detection is relevant to many areas of interest in computer science and engineering among them being, human-computer interaction (HCI), security, and health. The problem with traditional approaches that use fixed sets of features is that facial data may be highly variable and the approaches may then demonstrate poor reliability. In this study, we present a novel approach for the classification of prompted emotive voices which employs a convolutional neural network (CNN) for feature extraction and a support vector machine (SVM) for classification. CNN-SVM hybrid is a technique that combines the advantages of the CNN and the SVM to improve accuracy and reduce the level of random variation. Using the abovementioned proposed method, the system was tested on a benchmark data set to get an accuracy of 0.92 accuracy which surpasses standalone CNN and SVM results as well as other recent works on a similar problem. The findings presented here show that the proposed CNN-SVM fusion model can be used to successfully identify emotions from facial expressions, especially for real-life applications. Future work will investigate additional possibilities with superior model learning and increase the scope of the data used.