Improved area-based stereo matching using an image segmentation approach for 3-D facial imaging

Orthodontists and cranio-facial surgeons are particularly interested in assessing in a qualitative and quantitative manner changes in facial morphology. We describe an improved intensity-based matching algorithm for convergent stereo camera axes which has been optimized for recovering a dense map of facial surface points. Various refinements have been incorporated into the matching strategy including image segmentation, scale adjustment for correction of perspective distortion, relaxation, bi-directional matching and texture projection. It has not been necessary to perform an initial rectification stage. Some preliminary results are presented on the accuracy of metric reconstruction using a calibration object and a qualitative study using a recovered 3-D facial model of a human subject. We then evaluate the potential for 3-D image registration as a first step towards the detection of facial shape change.