Facial expression recognition with optimum accuracy based on Gabor filters and geometric features

In this paper a new method for facial expression recognition is presented. According to this algorithm, an appropriate mask is designed using Gabor filters, and it is convolved with original image. Then oval part of face is specified by using PZMI and its main components, such as eyes, mouth, eyebrows and nose, are characterized. Then defined points and distances are selected on these components automatically. This work is implemented by comparing positions of defined points (land-marks), and lengths of distances with their values in normal facial expression, and using a RBF neural network as the classifier. Finally results are presented. This method has high accuracy in comparison with other methods.