Spontaneous facial expression analysis using optical flow

Abstract

Investigation of emotions manifested through facial expressions has valuable applications in predictive behavioural studies. This has piqued interest towards developing intelligent visual surveillance using facial expression analysis coupled with Closed Circuit Television (CCTV). However, a facial recognition program tailored to evaluating facial behaviour for forensic and security purposes can be met if patterns of emotions in general can be detected. The present study assesses whether emotional expression derived from frontal or profile views of the face can be used to determine differences between three emotions: Amusement, Sadness and Fear using the optical flow technique. Analysis was in the form of emotion maps constructed from feature vectors obtained from using the Lucas-Kanade implementation of optical flow. These feature vectors were selected as inputs for classification. It was anticipated that the findings would assist in improving the optical flow algorithm for feature extraction. However, further data analyses are necessary to confirm if different types of emotion can be identified clearly using optical flow or other such techniques.