Emotion Modelling Using Neural Network

• Main Author: Lam, Choong Kee
• Format: Thesis
• Language: English; English
• Published: 2005
• Subjects: QA71-90 Instruments and machines
• Online Access: http://etd.uum.edu.my/1252/1/LAM_CHOONG_KEE.pdf; http://etd.uum.edu.my/1252/2/1.LAM_CHOONG_KEE.pdf; http://etd.uum.edu.my/1252/

Emotion has become an important interface for the communication between human and machine. Human's emotion can be detected by the machine, and machine can respond to it and interact with human in a more natural and adaptive environment. This study attempts to model emotion using neural network technique. Six primary emotions considered in this study are anger, disgust, fear, happiness, sadness and surprise. For data preparation, front views of child facial expression images have been captured with Sony Cybershot DSC U50 digital camera and extrated using MATLAB Image Processing toolbox. A dataset consists of 120 patterns with 82 attributes and emotion targets have been gathered at the end of image processing activity. The dataset was tested on Multipayer Perceptron with backpropagation learning algorithm. The emotion model obtained in this study uses parameters such as; learning rate 0.1, momentum rate 0.1, Sigmoid activation function, 200 epoch learning stopping criteria, with its architecture, 82 input units, 10 hidden units and 6 output layer units. The Neural Network performance achieved 97.50 percent accuracy whereas the regression model obtained 66.67 percent accuracy. This result indicates that neural network has high potential to be used as emotion.