Customer satisfaction (CS) measuring has become one of the strategic tools for companies. Many methods exist to measure customer's satisfaction. However, generally results could be wrong. Otherwise, CS can be deduced from customers emotion state. In this paper, we propose a new end-to-end method for facial emotion detection. For that, six new characteristic features are proposed. We characterize especially the most significant emotions namely "Happy", "Surprised" and "Neutral". The proposed method is invariant to camera position. A very challenging datasets as Radboud Faces and Cohn-Kanade (CK+) are used for study and evaluation. Obtained results show that our method reaches a high recognition accuracy and outperforms Action Unit (AU) features based Support Vector Machine (SVM) and K-Nearest Neighbors (KNN).