Emotion Recognition Using Facial Images

An important component of human life and behaviour is emotion. Automatic emotion recognition has grown in importance in the domains of affective ciphering and machine-human interaction in recent years. Among the many physiological and active indicators that could be used to identify emotions, taking pictures of facial expressions is one of the most straightforward and affordable methods. Due to architectural, cultural, and environmental variations, developing a universal, cross-subject model for emotion identification from facial expressions is currently difficult. On the other hand, utilizing conventional machine learning techniques would necessitate a sizable collection of labelled examples in order to develop a subject-specific, personal model. Due to these factors, we suggest using transfer learning in this work to create domain-specific models for extracting the valence/arousal dimensions of face image emotional content. Transfer learning enables us to apply the feature extraction functionality in the single-subject situation while reusing the knowledge acquired from a huge multi-subject dataset by a deep convolutional neural network. By doing this, it is possible to use less labelled data when training a customized model than if only subjective data were used.