Facial expression recognition

Abstract

The main focus of the project focused on recognising human facial expressions given an image. The output of the task would accurately detect the given emotion from the image using a carefully built algorithm. Facial Expression Recognition (FER) is used in practicality to reveal the emotional response of a human in the absence of a verbal setting. The area of Facial Expression Recognition has been approached in both traditional image classification algorithmic methods as well as Artificial Intelligence (Deep Learning) in the past few decades. While traditional methods still prevail largely, Deep Learning has been employed to improve efficiency and precision. In this course of the project, Convolutional Neural Networks has been experimented to achieve a result of appropriate efficacy. The algorithm aims to identify 6 different kinds of expressions namely, happiness, anger, sadness, fear, disgust and surprise. The project made use of the platform of Python programming language, deep learning frameworks such as Tensorflow, OpenCV and Keras. The training set used as the base input is the FER2013 dataset with 35,888 images which had been further split in the ratio of 80:20 into training and test set which was inputted into the neural network to accomplish the operation and produce the result. The facial features were learnt by the built network at each stage and with the help of the neural networks, the expected emotion could be identified with accuracy.