FairEmotion

Emotion classification, also known as facial expression recognition (FER), is a growing area of research interest in computer vision. It consists of three major steps: face detection, feature extraction, and emotion classification. Emotion recognition could be applied to many fields including virtual reality (VR), human-computer interaction (HCI), and mental health. Following the rise of deep learning, deep neural networks have been applied to emotion recognition, which reduces the dependence on physics-based models and image pre-processing. One of the biggest challenges of emotion recognition is performance biases across racial groups. Potential sources of the biases include aggregation bias, measurement bias, and unbalanced training data. Fairness in emotion classification is crucial for avoiding discrimination and ensuring equal opportunities across different racial groups.