FACIAL EMOTION RECOGNITION USING DEEP CONVOLUTIONAL NETWORKS

Abstract

Facial emotion recognition is a developing area that is used in a variety of modern applications such as social robots, neuromarketing, and gaming. Nonverbal communication methods like as facial expressions, eye movement, and gestures are widely employed in various applications of human computer interaction, with facial emotion being the most commonly used since it conveys people's emotional states and sentiments. Emotion recognition is a difficult endeavour since there is no landmark demarcation between the emotions on the face, and there is also a great deal of complexity and unpredictability. Traditional machine learning algorithms employ certain crucial extracted characteristics for modelling the face, therefore it cannot reach high accuracy rates for emotion identification since the features are hand-engineered and rely on existing knowledge. In this work, convolutional neural networks (CNN) were constructed to recognise facial emotion expressions and classify them into seven fundamental categories. CNN calculates features by learning automatically rather than by hand-engineering them. The suggested approach is unique in that it uses facial action units (AUs) of the face, which are first recognised by C NN and then used to recognise the seven fundamental emotion states. The Cohn-Kanade database is used to analyse the proposed model, and the model obtains the best accuracy rate of 97.01 by adding AU, whereas other efforts in the literature employ a direct CNN and reach an accuracy rate of 95.75.