A method for facial expression recognition for a speaker was proposed using thermal image processing. We reported that we got the average recognition accuracy of 90% for one speaker, when she exhibited one of the intentional facial expressions of “angry”, “happy”, “neutral”, “sad”, and “surprise”. In this study, we improved our system to be able to save thermal static images automatically at the times decided to recognize facial expression for a speaker with or without glasses. In this method, the above five kinds of facial expressions were discriminable with the average recognition accuracy of 88% for seven speakers. Then, we analyzed the influences of individual variations on facial expression recognition.
View full abstract