Automated Feedback Generation System using Facial Emotion Recognition

Advertisements have come a long way from textual content to graphical content and finally video advertising, where an advertisement video is played before or while the user is viewing any content of his interest on the internet. Feedback for an advertisement or a service is essential for its improvement which is often taken from the user through forms. Instead of asking to fill up forms, here facial expression analysis can be used to generate the feedback without having the user intervention. For the purpose of emotion detection, a Support Vector Machine has been employed to detect six types of emotions namely - happy,sad,anger,surprise,disgust and neutral. First the face is detected and 68 facial landmarks co-ordinates are located on the face that correspond to eyes, mouth, nose and jaw. With this landmarks, a total of 15 real valued parameters are calculated taking into account the neutral face of the user. These real valued parameters are computed as ratios of different distant metrics corresponding to the actual face and neutral face. Experimental results show that an accuracy of 90.60 ± 1.65% has been achieved in recognizing the six emotions by using 20% of dataset for testing. An average accuracy of 88.06% is obtained in 10-fold cross validation testing. This information is used to generate a visual feedback to the advertiser. Apart from generating feedback for an advertisement, this can also be used to get opinions of people wherever communication or language becomes a hindrance either due to illiteracy or not knowing the language.