Avaliação de classificadores HAAR projetados para detecção de faces

Abstract

This dissertation work presents the development of a new evaluation method for Haar classifiers designed to locate faces in images, in order to help researchers choose the best classifier for their own need, as well as to propose a new metric to evaluate future classifiers designed for face location. Face location, the primary step of the vision-based automated systems, finds the face area in the input image. An accurate location of the face is still a challenging task, but it enables a more efficient segmentation, making the identification and recognition of individuals more accurate. Viola-Jones framework has been widely used by researchers in order to detect the location of faces and objects in a given image. Face detection classifiers, used by Viola-Jones framework, are shared by the scientific and academic community. Nevertheless, few works have studied their accuracy. By applying 10 of these classifiers in two different face databases (FEI and Yale), an analysis was made and a new heuristic was created considering scores given to 22 facial landmarks. The results appear to be more accurate than the ones obtained by the only heuristic proposed by another work, developed so far.