Abstract:

Facial expressions are an important part of human-machine interaction as well as playing a substantial role in human communication. Since facial expressions are significant parameters in decision making on such important issues as criminal detection, monitoring the attention of the driver and patient follow-up, automatic detection of facial expressions via systems has become a popular subject. In this study, the detection of facial expressions is aimed by selecting the robust features affecting the detection of emotions in facial images. Face++ SDK is used for locating the facial keypoints. All probable distance data between the points and ratio data between the lines have been calculated, then the robust distance, ratio and distance + ratio features affecting the expression detection has been selected by Sequential Forward Selection (SFS) method. Following this step, each robust feature vectors has been classified and their success rates have been compared. For classification, Support Vector Machines (SVM) and k-Nearest Neighbors (KNN) are used. As a result, 91% success rate has been achieved with 4 robust ratio features on SVM. In our study, in which neutral, surprised, sadness, angry, happy facial expressions have been analysed, surprised has been predicted right by 100%, happy and angry by 95%.

Keywords: