An investigation on the type of data for facial expression recognition

Emotion is a fundamental component of being human. In human daily social life, knowing the emotional feeling of a person is intuitive but for a computer emotion recognition is harder. Emotion could be expressed through many social behaviours, including voice, gesture, psychological signals, etc., but the main part of the overall impression is the facial expression. FER (Facial Expression Recognition) deals with the automatic analysis of facial behaviours to recognize emotional states and, during the last decade, it found application in several fields, as human-computer interaction (HCI), virtual reality, advanced driver assistant systems (ADASs) and entertainment. Although interest in automatic facial expression recognition has been increasing and many resources are invested, FER still presents many challenges to tackle, one of these are databases. Databases play a fundamental role for the progress of FER techniques as they allow artificial intelligence to train and learn from their features. In order to have a useful database it is not sufficient collecting enough data, but they must also be processed to have the right information to give to the neural network. This work includes an investigation of the data exploited for facial expression recognition, procedures of acquisition and processing data, and state-of-the-art database testing.