Research On Automatic Registration Of TPS Based On Three-dimensional Skull And Facial Data

Three-dimensional craniofacial registration is to find the corresponding relationship among three-dimensional craniofacial samples of different sizes and different poses under a unified scale and coordinate system to achieve one-to-one correspondence.The current craniofacial registration methods relie on the feature points of manual calibration,which is time-consuming,labor-intensive and difficult to guarantee accuracy.For the abovementioned problems,automatical locationd of the craniofacial geometric feature points by extracting craniofacial radial curves or geodesics are proposed in this thesis,and the TPS automatic registration methods of skull and facial data are realized,which are automatic registration methods of skull and facial data with improved registration accuracy.The main research work of this thesis includes:1.3D craniofacial automatic registration methodFor 3D craniofacial data,two TPS craniofacial automatic registration methods based on radial curve and based on geodesic are proposed.Firstly,the radial curves and geodesics on the reference face are extracted,the inflection point on the radial curve or geodesics are calculated as the feature point on the reference face.Then their corresponding points on the radial curves or geodesics on target face are found as the target facial feature points through the consistency point drift(CPD)algorithm.Using the two sets of feature points as the control points of TPS transformation,the reference skin is transformed to the target skin by TPS.Finally,a registered skin is obtained by finding the closest point of the reference facial skin on the target facial skin based on the Qhull method.Comparing our method with the registration results of other face registration methods,such as ICP,random point TPS,and deformation map,experimental results show that our registration results are more accurate and more stable than other methods.2.3D skull automatic registration method.For 3D skull data with complex topology,an automatic TPS skull registration method based on radial curves and boundary features is proposed.The control points of this method are divided into two parts.The first part of the control points are the feature points on the critical parts of the mouth,eyes and nose on the radial curves which are extracted on the reference skull and the target skull.And the second part of the control points are the characteristic point of the upper hole on the reference skull and the target skull.Taking these two sets of feature points as the control points of TPS transformation,the reference skull is transformed to the target skull by TPS.Finally,a registered skull is obtained by finding the closest point of the reference skull on the target skull based on the Qhull method.Compared with random feature points TPS registration method,experimental results show that our registration results are more accurate and stable than other methods.3.A craniofacial data registration systemBased on above researches,we have developed a three-dimensional craniofacial registration system based on curves,including different registration methods based on radial curves and geodesics.The interface is friendly and the operation is simple,which is convenient for users to carry out craniofacial registration.Based on our research,we can establish the accurate one-to-one correspondence on 3D craniofacial data,which can lay the foundation for the craniofacial reconstruction,3D face recognition and analysis,and promote the development of craniofacial morphology,forensic anthropology and etc.It also has important application value in public security criminal investigation,medical plastic surgery,archaeology and other fields.