No abstract

Victims of fetal alcohol syndrome (FAS) exhibit a unique facial phenotype that is emphasised in diagnosis. Among the characteristic facial features, several occurring in the region of the orbits can be evaluated quantitatively using distance measurements. An algorithm is described for automatic extraction and measurement of eye features from stereo photographs. The algorithm was applied to photographs of 46 six-seven-year-old children. The approach relies on peak and valley maps and integral projection functions, to locate the eyes and extract the iris, and genetic algorithms to fit cubic splines to the upper and lower eyelids. Measurements obtained automatically using this algorithm were compared with measurements obtained manually from the photographs. Mean absolute differences between automatic and manual measurements were less than 1 mm for palpebral fissure length (PFL) and interpupillary distance (IPD). Absolute differences were less than 1 mm for 80.4% of PFLs and 100% of IPDs. Inner canthal distance and outer canthal distance did not compare favourably with manual measurements.

Conventional screening for Fetal Alcohol Syndrome in children involves time-consuming and intrusive facial distance measurements. We are developing an alternative method that will allow automatic measurement of the relevant distances from a pair of stereo photographs. This paper describes a method for automatic extraction from such photographs of the points on the eyes used in the distance measurements, using peak and valley maps of face images.

Abstract-Conventional screening for Fetal Alcohol Syndrome in children involves time-consuming and intrusive facial measurements. We are developing an alternative method that will allow automatic measurement of the relevant points from a pair of stereo photographs. This paper describes a method for automatic eye extraction from such photographs, using genetic algorithms to match eye templates to face images.