Analysis of Facial Emotion Recognition for Image and Video Data using Convolution Neural Networks

Facial Emotion Recognition is used to classify the emotional states of a human. The objective is to group every facial picture into one of the seven facial emotion classes like Surprise, Sad, Neutral, Happy, Fear, Disgust and Angry. Convolutional Neural Network (CNN) is used for classifying the emotion. Various gray scale pictures from the dataset and real-time videos are taken as input. Then feature extraction is done using the series of convolution and pooling layers of CNN and classification is done using softmax layer. To overcome the overfitting problem of the model, various procedures including dropout, cluster standardization and L2 regularization are used. Experiments are performed on Facial expression dataset image folders (fer2013) dataset and achieved better accuracy with our model than the existing works in predicting the individual emotions. In addition to this, the model is also used to predict the emotion of each image in the real time video data with good performance.

I. Introduction

Emotion recognition is likely to produce the best results when multiple modalities such as text, sound, image, and video are used to understand a specific human being's behavior/feelings. Recognition of emotion gives advantages to numerous establishments and parts of life. It is helpful and significant for security and medical services purposes. Additionally, it is essential for simple and straightforward recognition of human physiology at a particular instant without really asking them [1]. Emotions are communicated while collaborating and associating with others. Emotions are generally classified as happy, surprise, neural, fear, sad, anger.