Face Reconstructions using Flesh Deformation Modes.pdfVIP

Face Reconstructions using Flesh Deformation Modes Peter Tu1, Richard I. Hartley1, William E. Lorensen1, Majeid Allyassin1, Rajiv Gupta2, Linda Heier 2 1 GE Corporate RD Center 1 Research Circle Niskayuna, N.Y. 12309 2 Weill Medical College Cornell Cornell University Abstract An important forensic problem is the identification of human skeletal remains. For this purpose the skull is commonly used as a means for building a face model, which may be used for recognition of the subject. This paper describes a method whereby a 3D computer model of the face and head is built using computer 3D graphics techniques. In reconstructing the shape of the face from a skull, it is essential to have data relating the shape of the face to the shape of the underlying skull. In manual reconstruction techniques this information is derived from flesh depth tables. We offer a method in which Computer Tomography (CT) scans are used to get a dense set of measurements of flesh depth. To model the variation of the facial shape, Principal Component Analysis is used as a means of exploring the space of the major facial deformations. After a subject skull and a set of database skulls have been aligned, the average face and principal deformations may be computed. Also provided are interactive tools for substituting facial features (such as nose, eyes, ears) from a catalog of such parts. 1 Introduction In this paper, a database of CT (Xray Computer Tomography) head scans is used to reconstruct the face of an unknown subject given its skeletal remains. Skull and face surfaces can be extracted from each CT head scan. By establishing point correspondences between the CT skull and the subject’s skull, the CT face can be morphed to coincide with the subject’s skull. The morphing process removes differences that are caused by deviation in skull structure. Any remaining discrepancies can be attributed to variation in soft tissue. By morphing each head scan in the CT database, a collection of estimates o