Facial Emotion Recognition and Detection Using Convolutional Neural Networks

Facial emotion recognition is now becoming a hot topic in previous years as it helps in human computer interaction. As human emotions can be detected, there are various studies on it in the computer vision field. Humans can easily recognize the type of emotion from someone's facial expressions but it's tough for a computer machine. FER can be used in vast areas for applications such as education, social media, healthcare, Al based application, etc. In this research paper, the Convolutional Neural Network (CNN) technique has been built. The methodology is mainly divided into two steps: Haar cascade features from the face are extracted by using Viola Jones algorithm leading to facial identification and then facial emotion is recognised by using the help of neural network. This technique is built on conventional neural networks (CNN) and it has used FER2013 and CK+ dataset for detection of human emotions. This model consists of 3 completed connected layers and six convolutional layers. On the basis of our experimental results, this model has a validation accuracy of 68.26% with FER 2013 dataset and 91.58% with CK+ dataset.

I. Introduction

Emotion is considered as an important characteristic of human beings and they use it to cover their words directly. These emotions can be expressed through our expression of face, human made conversations, gestures and attitude too. Many individuals approach the next person on the basis of emotion that they make on their face. According to research done by the psychologists Mehrabian et al, facial expressions by humans are quite essential in human daily communication, with a transmission rate of 55%, far more significant than the speech (38%), language (7%), and so on [1]–[5].