54 An Adaptive Facial Feature Tracking for Expression Recognition Using Temporal Information in Feature Selection

In this work, more accurate facial expression recognition system is developed by proposing an enhanced algorithm in facial feature tracking and a novel feature extraction method. In this system, the selected facial feature points are tracked automatically from, the first frame to the last, using an enhanced cross correlation algorithm with adaptive window search. Based on action units, an appropriate feature selection differentiates between the expressions according to their temporal information. The use of principle component analysis reduces the less important information of feature matrix and accelerates the system's performance. The extracted feature matrix is then given to the K-nearest neighbor classifier with free parameter of K to classify the face expressions within the six basic emotions. The correct classification rate of this system is increased to 95.6%.