Emotion Recognition via Facial Expressions

For the last decade, a rising need for emotion recognition has been noticed in several domains, such as virtual reality, human-computer interaction, video games and health monitoring, etc. Effectively, emotion recognition via facial expressions attracts increasing attention. Based on geometrical facial features, this paper proposes a new facial emotion recognition method. We collected a novel dataset of 17 subjects facial performance of six emotional states (anger, fear, happiness, surprise, sadness, and neutral) using Kinect (v1) and Kinect (v2) and RGB HD camera. New positional features including a combination of angle and distance features are used to train the classifier. The K nearest neighbors (k-NN) is used as the main classification technique. To assess our proposed method performance, we use the leave-one-out subject cross-validation. A comparison between RGB and RGB-D data is provided. The obtained results show the superior performance of the RGB-D features provided by Kinect (v2). We observed in our experiment that the 2D images are not robust enough for facial emotion recognition due to the sensitivity of the RGB camera to the surrounding conditions.