A fuzzy approach to find Hirschberg points and to determine fixation in digital images of infants

Abstract

Screening infants for eye problems is loaded with uncertainty. Babies are unable to describe symptoms and in general may not be cooperative. In an ongoing research project, we are developing methods to screen infants for amblyopia, a common, but treatable, eye problem. The approach consists of processing a sequence of digital frames of the baby, searching for the few images where the infant "fixes" on a light positioned by the camera. Measurements made on the detected pupils are used to produce fuzzy confidence values that are fused together to create an overall confidence of fixation (the key factor in determining amblyopia). One of the most important and difficult factors in this calculation is the determination of the Hirschberg points - points of reflection of the light source off the front of the eye- if they exist at all. The criteria for detection are best thought of as fuzzy rules and methods to score potential Hirschberg points are developed. Results are shown on a variety of imagery collected in a clinical setting.