Reconhecimento de expressões faciais baseado em Active Appearance Model

Abstract

This work proposes the solution of a system capable of classifying, according to Paul Ekman’s principles, basic behavior expressions in a human face on a digital image. The first step refers to the outlining of the expression on the image. For this, a method based on the modeling algorithm AAM (Active Appearance Model) is applied. This method has an exceptional ability of adjusting a group of points whose shapes can be proper for a face. In order to initiate the classification process, the characteristics of the image with the defined expression, as well as the neutral expression one, are extracted, and the differences between them a classifier as Support Vector Machine and Neural Network performs the process of recognizing the expression. Finally, to authenticate the algorithm and the comparing tests, simulations with the facial expressions data bank, widely known as Cohn-Kanade (CK+), will be used. The results suggest that it is possible to associate muscular deformations, caused by facial expressions, with the Delaunay triangulation of landmarks reached by fitting AAM technique.