Abstract
A method to analyze three dimensional structure of a face using a pair of stereoscopic images of a face is proposed. Inputted stereoscopic images are compared each other to get corresponding point pairs by a kind of relaxation matching method. Obtained three dimensional data of the face are represented by patches defined by B-Spline function, and curvatures of these patches are used to get the top point of a nose. Three dimensional data are also used to decide potential positions of eyes, eyebrows and a mouth, and their exact positions are obtained from pixel values of the inputted stereoscopic images.
