The recognition of emotions through the perception of facial expressions can have several study applications, with applications in sociology, anthropology, and, especially, in clinical and educational areas. The interpretation of human emotions is a new horizon of understanding that is opened by understanding the ways people communicate and the real meaning of non-verbal behavior. The use, in literature, of controlled images considering human emotions datasets makes the proposed approaches effective indeed, but very restricted to this specific domain and controlled environments. This article aims the emotion evaluation and classification for more realistic means such as the self-acquisition of images with a webcam from a personal computer. In sum, validate a method classification developed on previous studies by this research team from facial coordinates. The best architecture showed an average accuracy of 46.8% among all classes. The best and worst emotions classified are in accordance with the literature.