| dc.description.abstract | This work presents a system that automatically performs the inference of basic emotions (happiness, sadness, anger, fear, disgust and surprise) through facial expressions from a computer user, using images captured by a webcam. The developed application is based on facial coding system FACS, that classifies specific facial actions, known as AUs (Action Units). The proposed approach consists in collecting moviment data of mouth, eyes and eyebrows to classify, by neural networks, AUs codes presents in performed facial expressions. With decision tree, ruleset or neural network, the emotions of AUs, previously classified, are inferred. The designed system was evaluated in three different scenarios: (1) using samples of faces bases to evaluate the recognition of AUs and emotions; (2) with samples of face bases to evaluate emotion recognition by neural network (alternative approach); (3) using a sample of images captured by webcam for evaluation of emotion in decision tree and neural network. As results, was obtained 53.83% of recognition rate over AUs, which implicating 28.57% of emotions recognition with decision tree - Scenario 1. The emotion inference by neural network achieve, 63.33% of recognition rate as the best result - Scenarios 2 and 3. The developed paper can be used to adjust computer?s behavior to address user?s affective state, or provides data to other softwares, such as intelligent tutoring systems. | en |
