Eye detection using a deformable template in static images

In this paper, we present a methodology that uses a deformable template to detect the human eye in static images. Geometrically, the used template is represented by two distinct entities: a circumference, that defines the iris contour; and two parabolas, one concave and other convex, that define respectively, the above and below contours of the eye. The geometrical shape of the used template is controlled by a set of eleven control parameters that allows its change in scale, position and orientation. To process the matching between the eye, represented in the input image, and the template used, we consider information obtained from four energy fields. Then, to process the dynamic update of the template parameters, we use an energy function, based on those energy fields, that characterizes the cost of the template deformation. During each of the seven processing phases, we search iteratively for the best combination parameters values that minimize the energy cost of the template deformation. In this paper, we illustrate the functionality of this method by presenting some experimental results, and present some conclusions and perspectives of future work as well.