3D Feature Line Based Statistical Face Reconstruction

The aim of craniofacial reconstruction is to estimate a face outlookassociate with the shape of the skull.Traditional computer-aided techniquesmimicmanual work using single staticfacial template, or deforming a reference skull in order to match the remaining skull. The resulting reconstruction is biased by the specific choice of the template or the reference. Thereafter, statistical model is used to avoid such bias. However, statistical facial model often meets several problems. One of them is small-size samples, which remain essential and largely restrict the statistical model. Others like how to efficiently extract facial feature. In this dissertation, we propose a statistical facial model based on3D feature lines. It is the lines rather than whole face that be trained for further processing. Comparing to traditional statistical craniofacial model, newly defined model have many advantages.First, training data is reduced remarkably. Second,statistical model is more representative and more robust to mesh noise.Third, automatic extraction of feature lines is more efficient and accurate than mouse picking.The main contents of this dissertation are listed as follows, of which part Ⅱ and Ⅲ are the most essential:1) Pre-processing facial training set.2) Extracting facial3D feature lines.3) Statistical model based on3D feature lines, aiming for craniofacial reconstruction.4) FaceVis: A framework of facial reconstruction, development and implementation.