Stereo image matching for facial feature measurement to aid in fetal alcohol syndrome screening

The facial phenotype associated with fetal alcohol syndrome is identified through the measurement of facial distances and their comparison to population norms. As an alternative to time-consuming direct manual measurements, stereo photogrammetry has been used to obtain the required facial measurements from stereo digital photographs. While stereo photogrammetric measurement is faster, it requires manual marking of relevant facial landmarks on digital images. We present an algorithm that automatically finds matched feature points on the second of a pair of stereo images, after manual marking of the first. Standard image processing tools are used for preprocessing. Matching is based on a simple exhaustive search in an image window, with the sum of squared differences of the pixel intensities in the two images as the objective function. Eye measurements, namely palpebral fissure length, interpupillary distance, inner canthal distance and outer canthal distance, as well as distances that can be used to approximate the circularity of the upper lip, were obtained using the manual method of marking both images, and the method of automatic marking of the second image. Comparison revealed mean differences less than 1 mm.