NOVEL FACIAL FEATURE EXTRACTION TECHNIQUE FOR FACIAL EMOTION RECOGNITION SYSTEM BY USING DEPTH SENSOR

In this paper, the novel approach to extracting the facial features from the movement of the facial skeleton model is introduced. In this approach, the data streams, the sequences of the normalized Euclidean distance between pairwise points on the facial skeleton model, are analyzed by using the Structured Streaming Skeleton (SSS) method to construct the SSS feature vectors SSS method was firstly introduced in body gesture recognition system to handle the persistence of intra-class variations. The assessment of the system performance and accuracy was conducted by K-Nearest Neighbors (K -NN) and the Support Vector Machine (SVM). The fifteen participants' data set collected by our designed software was used in the experiment. By considering the facial emotions as facial gestures, SSS method was extended to handle the problems of intra-class variations in facial emotion recognition system. The present approach using K -NN attained a 91.17% +/-2.36% of accuracy rate, which was better than a 67.83% +/-4.54% of accuracy rate obtained by that using SVM. The comparison of the presented approach with the state-ofthe-art was limited due to the unavailability of their data set. It could be concluded that our approach has achieved superiority over previously reported approaches by overcoming the intra-class variations.