Human Emotions Recognition Based On Facial Expressions

Shamala Dewi, Mariappan (2020) Human Emotions Recognition Based On Facial Expressions. [Final Year Project Report] (Unpublished)

	PDF shamala (24 pgs).pdf Download (6MB)
	PDF (Please get the password from Technical & Digitization Management Unit, ext: 082-583913/ 082-583914) shamala (fulltext).pdf Restricted to Registered users only Download (30MB)

Abstract

Human emotion recognition is a widely researched topic. Studying facial expression plays a major role in emotional perception, contributing to the creation of the paradigm for humancomputer interaction (HCI) that can be applied in the fields of neurology, lying recognition, smart environments and paralinguistic communication. In this research project, three pretrained Convolutional Neural Networks were trained, tested and evaluated using MATLAB R2020a. Those networks are simple CNN net designed using 10 layers, AlexNet that consists of 25 layers and DenseNet-201 that consists of 708 layers. The dataset used to train and test the networks is an existing dataset called Bosphorus 3D Face Dataset. The model evaluated using K-Fold Validation and confusion matrix methods. The simple CNN net, AlexNet and DenseNet achieved an accuracy of 16.67%, 87.2% and 97.62% respectively. IO-Fold Validation test evaluated the DenseNet model with an accuracy of 83.33%. Confusion matrix gave the same accuracy for all the three networks. The high accuracy of the DenseNet has set a great benchmark for the future works of human emotion recognition. DenseNet is a good pretrained network that can be used for classifications.

Actions (For repository members only: login required)

View Item