DENSE CORRESPONDENCE AND STATISTICAL SHAPE RECONSTRUCTION OF FRACTURED, INCOMPLETE SKULLS

Abstract

Human skulls can be fractured or incomplete due to impact fracture or congenital defects. In this thesis, a statistical shape model for skulls is built to reconstruct the normal appearance of fractured, incomplete skulls. To build the statistical shape model, a dense correspondence algorithm is proposed to resample the skulls. With anatomical landmarks as hard constraints to ensure anatomically consistency, and control points sampled on the skull surfaces as soft constraints to provide additional local shape constraints to ensure close matching, the proposed dense correspondence algorithm is more accurate than existing methods. The error source of skull reconstruction is also investigated. The two primary sources of skull reconstruction error are resampling error and ASM fitting error. Test results also show that there is a linear relationship between ASM fitting error and reconstruction error, which allows reconstruction error of real fractured, incomplete skulls to be estimated by ASM fitting error.