Emotion recognition using soft computing techniques

• Main Author: Wang, Zeqing
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/78276
• Institution: Nanyang Technological University

In the development of Artificial Intelligence, emotion recognition is an important aspect of computer-human interactions. The need for an accurate and robust emotion recognition system is especially important for robotic applications where the hardware resources are limited. In this work, an emotion recognition system based on facial expressions will be developed using soft computing techniques. The work examines the recent advances in deep learning in order to develop a convolutional neural network for the classification of facial expressions into 7 emotional states: Angry, Disgust, Fear, Happy, Neutral, Sad, Surprise. The model constructed is based on the Xception Network, utilizing inception modules residual connections, and depthwise separable convolutions. The network will be trained on images from the FER2013, Extended Cohn Kanade and Japanese Female Facial Expressions Datasets. The data are preprocessed by dectecting facial region, cropping/ resizing and histogram equalization before feeding into the network for training. Two models, a Convolutional Neural Network using sequential layers, and Principal Component Analysis with SVM (PCA+SVM)), are also created for comparison of accuracies and performance benchmark. From the results, it can be seen that the Xception model performed better than LBPH with KNN, but comparable to the sequential CNN. After tuning the model, the performance of the Xception Model surpassed sequential CNN. Emotion recognition is carried out using webcam and the predictions were made almost in real-time. Future direction in this work includes gathering more facial data that are specific to the demographics of the end users, using specialized hardware (e.g. Tensor Processing Unit) and camera achieve better performance.