Craniofacial image analysis is traditionally performed with a set of manually obtained landmarks of the face. Our work, on the contrary, uses full meshes of the subject's head to extract image features that lead to classification, quantification, and image retrieval. This talk will discuss two different but related studies. The first uses histograms of the azimuth and elevation angles of surface normals of the mesh representation of craniofacial anatomy in order to classify and quantify shape abnormalities. The second uses local curvature features to learn the characteristics of regions about facial landmarks and then learns which of these can be successfully used to find the center plane of the human face, even in the presence of clefts. Experiments on three different abnormal conditions are described.
Wednesday, January 30, 2013
Free and open to the public