Reconhecimento automático de expressões faciais baseado em modelagem estatística

• Ano de defesa: 2013
• Autor(a) principal: Pedroso, Felipe José Coelho
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Processamento de imagens; Interação homem-máquina; Sistemas de reconhecimento de padrões; Expressão facial; Aprendizado do computador; Engenharia Elétrica; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/9624

Facial expressions are constant targets of studies since Charles Darwin in 1872. Research in psychology and highlighted the work of Paul Ekman claim that there are universal basic facial expressions and they are expressed in all human beings regardless of factors such as gender, age, culture and social environment. Although you can create new more complex expressions combining the fundamental expressions of happiness, sadness, fear, disgust, anger, surprise and contempt, beyond the neutral face. The matter is still relevant, since there is a great need to implement human machine interfaces (HMI) able to identify the expression of an individual and assign an output consistent with the observed situation. One can cite as examples iterations man-robot surveillance and motion graphics. In this work it’s proposed an automatic system to identify facial expressions. The system is divided into three blocks: face localization, feature extraction and identification of facial expression. The Japanese Facial Expression Database - JAFFE was used for training and testing. The location of the face is done automatically using the framework proposed by Viola and Jones estimating center of the face. Following the Active Appearance Model - AAM algorithm is used to describe statistical model of shape and texture to the database. With this descriptor is possible to generate a vector capable of representing faces with reduced dimension and hence the facial expression contained therein through an iterative search algorithm from an average model. This vector is used in recognizing facial expressions block, where the classifiers are tested based on the nearest neighbor k-NN and support vector machine - SVM with RBF kernel to address the problem of non-linear way. A mechanism to decrease the error of the model is proposed before the output of the face detection block, because the success of the algorithm is highly dependent on the starting point of the search. A change in the AAM algorithm is also proposed to reduce the convergence error between actual and synthetic model that is addressing the problem of nonlinear way. Tests were conducted using leave one out cross validation for all the facial expressions and the final classifier was SVM-RBF. The system has an accuracy rate of 55.4%, with 60,25% sensitivity and 93,95% specificity.