| dc.description.abstract | Facial expressions are the most powerful and natural means of communication among human beings. The face can express emotion sooner than people verbalize or even realize their feelings. By knowing the effectiveness of facial expressions in daily life, recently many researchers have attempted to recognize the facial expression due to its potential applications in behavioural science, medicine, security, biometrics and human machine interaction. Although many researchers have proposed various methods to detect and recognize facial expressions, this field still remains very challenging especially in real time applications. In this study, luminance stickers based facial expression recognition is proposed. A set of minute luminance stickers are fixed on selected locations in a subject’s face and the expressions are captured by video recording. Face feature variations are identified through the movements of the luminance stickers which represent the facial muscular activities. The recorded videos are processed using some software to extract the 2D coordinate values of each sticker from each frame. This set of 2D coordinates vectors are used as an input data for feature extraction method to recognize facial expressions. In this research, facial expressions are recognized using three feature extraction methods such as AR Model, FFT and DWT with their different orders to study their effectiveness. Six statistical measures such as standard deviation, variance, mean, energy, power and entropy are computed for the extracted feature coefficients of AR Model, FFT and DWT to select the best statistical measure which contributes to higher accuracy of facial expression classification. Conventional validation and cross validation are performed on the selected statistical measures to study the reliability of the classifiers. Three different classifiers namely ANN, kNN and LDA are employed in this study to investigate their performance in correctly classifying a total of eight facial expressions. From this analysis, it is found that the facial expression recognition using DWT gives very promising accuracies ( between 98.83% and 99.15% ) of facial expression classification for kNN and LDA classifiers compared to other feature extraction methods and classifiers. It is, hence, concluded that the proposed method can be used as a valuable tool for facial expression based applications such as human machine interfaces. | en_US |
