Aid for atypical individuals to understand emotions through a facial emotion recognition system

Abstract

Facial expressions in atypical persons, such as autistic individuals, differ from those in regular people. They might be of poorer quality or less precisely expressed. It is difficult for people to recognize emotions of such people and vice versa. Lack of sufficient sample data and access to personal information due to medical ethics imposes a constraint in understanding the pattern of facial expression in differently abled people and its recognition by common individuals. It may be easier to address this problem in the reverse direction. We aim to develop an interactive software to help atypical people recognize common facial expressions, and we think it will be beneficial to both types. This method may be used to any other atypical/typical people, as teaching generic expressions to a group of people is easier than comprehending the variations in each person's facial expressions individually. We develop a software based on a neural network that is trained on a dataset of images of facial expressions that are labeled with associated emotions. The trained network can then classify accurate emotions in real time for any new input images. Despite the network's simplicity and limited dataset size, the accuracy of the predictions obtained is rather good, which reduces training time and allows the network to function instantly. Real-time image capture and label display may also be achieved by integrating the program with a hardware device.