Judgments Of Emotional Reactions By Facial Emotion Recognition System: A Comparison

Abstract

The research begins by asserting that a change in the movement of the muscle group responsible for executing a facial action unit permits one to ascertain the person?s emotional state. The connection between muscle movements and emotions is what makes it possible to build up a recognition system. Emotion has physical correlates that are independent of race, culture, and age. We looked at how two systems recognize emotions by watching videos of people showing different emotions, both real and imagined. We also looked at videos of people who tend to be good at controlling their emotions, like politicians and leaders. Furthermore, we demonstrate that there is a difference in the emotion judgments by two major emotion recognition systems, Emotient and Affectiva, in both posed, spontaneous, and semi-spontaneous emotions, which we have traced down to the level of action unit in that emotion measurements can be correlated to the difference in the measurements of action units. This can be attributed to the difference in algorithms of the two systems. Furthermore, Posed dataset Ravdess and Spontaneous Dataset (AM-FED) and semi-spontaneous datasets from our collection (Politicians and Governors) baselines were used. Can emotion recognition systems with different architectures, training, and testing methods have consistent emotion detection results? The reliability and variation of the emotion recognition results on the spontaneous, semi-spontaneous, and posed database were examined using statistical techniques, such as the Spearman correlation coefficient, Kruskal-Wallis tests, Chi-square tests, and Pearson tests.