An image processing system for the automatic extraction of craniofacial landmarks

A novel algorithm has been developed to locate craniofacial landmarks on skull X-rays. The locations of these landmarks are used by orthodontists for cephalometric evaluations. The proposed image processing method uses mathematical morphology to generate a target recognition error function whose surface is minimized to estimate the probable location of a landmark. An automatic training procedure accompanies the proposed algorithm. The training provides the means by which the algorithm may automatically learn to deal with subtle differences in a landmark's appearance. A novel decomposition technique that provides some size invariance and a method for minimizing the search window sizes are also presented. A system was trained to locate 20 landmarks. Tests on 40 X-rays showed an 88% recognition rate and a mean location error of 1.1 mm.<>