RF-NFN: Residual Neuro-Fuzzy Network-Based Multi-Modal Facial Expression Recognition

Abstract

Human beings show multiple responses to various emotional states, like anger, disgust, surprise, happiness, sadness, and fear. Among various emotions, facial expressions are widely informative as they exhibit a person's intentions and character. Facial expression recognition is used in many applications, such as marketing, research, customer service, neuroscience, and psychology. Traditional unimodal methods for facial expression recognition are ineffective due to the scarcity of data. In this paper, the Residual Fused Neuro-Fuzzy Network (RF-NFN) is used for the recognition of facial expressions and detection of emotion type using video and Electroencephalogram (EEG) signals. Here, the video frame is allowed for pre-processing done by Non-Local Means (NLM) filtering. Then, pre-processed video frames and input EEG signals are fed toward feature extraction, which is then followed by feature selection. Finally, facial expressions are recognized and the type of emotion is detected by RF-NFN, which is designed by incorporation of Hybrid Cascade Neuro-Fuzzy Network (Hybrid Cascade NFN) and Deep Residual Network (DRN). Moreover, the performance of the RF-NFN model is validated by three performance measures that exhibited a maximum accuracy of 90.88%, precision of 91.77%, and recall of 94.57%.