Assessment of automatic facial expressions recognition "in the wild": a time-series analysis using GAMM and SiZer methods

The analysis of facial expressions is currently a favored method of inferring experienced emotion, and consequently significant efforts are currently be-ing made to develop improved facial expression recognition techniques. Among these new techniques, those which allow the automatic recognition of facial expression appear to be most promising. This paper presents a new method of facial expression analysis with a focus on the continuous evolu-tion of emotions using Generalized Additive Mixed Models and Significant Zero Crossing of the Derivative (SiZer). The time-series analysis of the emo-tions experienced by participants watching a series of three different online videos suggests that analysis of facial expressions at the overall level may lead to misinterpretation of the emotional experience whereas non-linear analysis allows the significant expressive sequences to be identified.