Evaluation of a System for Real-Time Valence Assessment of Spontaneous Facial Expressions

Over the last years many solutions have been proposed for facial expression analyses for emotion assessment, some of which claim high recognition accuracy in real time, even on natural expressions. However, in most (if not all) cases, these studies do not use a publically available and validated dataset for inducing emotions and fail to place the reported performance in the right context by comparing against human expert scoring performance. We evaluate the abilities of FaceReader, a commercially available product for fully automatic real-time emotional expression recognition, to detect spontaneously induced emotions. We presented a balanced subset of IAPS pictures to a number of test subjects and compared the automatically detected emotional valence as reported by FaceReader to the IAPS valence labels and to the evaluation of three psychologists.