Publication Date

Summer 6-21-2021

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)


Computer Science

First Advisor

Nada Attar


Facial Expression Recognition, Image pre-processing, Deep Learning, Transfer Learning, Convolution Neural Network [CNN].


Humans often use facial expressions along with words in order to communicate effectively. There has been extensive study of how we can classify facial emotion with computer vision methodologies. These have had varying levels of success given challenges and the limitations of databases, such as static data or facial capture in non-real environments. Given this, we believe that new preprocessing techniques are required to improve the accuracy of facial detection models. In this paper, we propose a new yet simple method for facial expression recognition that enhances accuracy. We conducted our experiments on the FER-2013 dataset that contains static facial images. We utilized Unsharp Mask and Histogram equalization to emphasize texture and details of the images. We implemented Convolution Neural Networks [CNNs] to classify the images into 7 different facial expressions, yielding an accuracy of 69.46% on the test set. We also employed pre-trained models such as Resnet-50, Senet-50, VGG16, and FaceNet, and applied transfer learning to achieve an accuracy of 76.01% using an ensemble of seven models.