Multi-dimensional emotion recognition from geometry and color information
Emotions play a fundamental role in everyday interactions among humans. Humans are adept at expressing themselves and interpreting others through a multi-modal, subtle and complex process using non-verbal cues including speech prosody, facial expression, eye gaze, body gestures, head motion, posture, and skin color changes. However, recognizing the affective state of humans is a difficult task for computers. Automatic emotion recognition has focused on analysis of the six discrete basic emotions: happiness, sadness, surprise, fear, anger and disgust. However, humans express more complex and subtle affective states such as confusion, shame, pleasure, anxiety or depression. Therefore, a different representation based on a small number of continuous latent dimensions is more suitable for emotion recognition. In this dissertation I explored the problem of multi-dimensional emotion recognition. The first step for gathering visual features is face detection; I developed an approach for face detection based on a new set of Haar features. The approach also includes a method based on a genetic algorithm to reduce the training time. According to my experiments, these new set of Haar features can attain similar results to other methods while generating simpler classifiers with fewer Haar features. I developed an emotion recognition approach based on a small set of high-level features instead of a large set of low-level features. First, I present a set of experiments with high-level features obtained from geometric information. These features include gaze direction, head tilt, smile level, and eyebrows interaction. The experimental results show an improvement over current approaches. Later, I used features from color information that measure changes in color for three face regions. My results show that facial skin color changes can be used to infer the emotional state of a person. Finally, I used a combination of features from geometric and color information that improved performance in emotion recognition.
Garcia, Geovany Abisai Ramirez, "Multi-dimensional emotion recognition from geometry and color information" (2014). ETD Collection for University of Texas, El Paso. AAI3623456.