3D Faces’Partitions Registration And Point Correspondence Algorithm

Currently, computer-assisted techniques to recover the skull face have been used more and more widely in archeology, forensic science, forensic science, orthopedic medicine and so on. Among all computer-aided craniofacial restoration techniques, comparing with other methods, the technique based on statistical theory can make the restoration results more scientific, and it has become a hot research. Craniofacial restoration techniques based on the statistical theory need a large number of training samples for statistical training, and the priori knowledge acquired will be used to guide craniofacial restoration. Accuracy point correspondence between training samples should be established, the accuracy of craniofacial restoration is largely dependent on point correspondence between face samples. Aiming at point correspondence between the three-dimensional models, this paper researches on point correspondence between3D face models. The main work and progress are listed as follow:1. Introduce the MMP exact geodesic algorithm to achieve the partition of the facial model. According facial feature points, define a set of facial segmentation templates based on features points. Use MMP algorithm to obtain geodesic, then according to the geodesic path, use the active point moves cutting triangulation algorithm to divide triangular mesh. This method makes partition more precise and boundary smoother.2. Facial partition registration algorithm based on a combination of ICP and TPS is proposed. Craniofacial registration consists of two steps:rigid registration based on ICP and non-rigid registration based on the TPS. ICP algorithm is a relatively common rigid registration algorithm, which can achieve rigid registration between two point sets. TPS is a global, non-rigid registration algorithm, which can enable data smooth after registration. Models after registration can be approximated coincide, which is helpful to improve the accuracy of the corresponding point.3. Introduce the multi-scale constraint based on the volume integral invariants and the method PCDM of3D point correspondence based on non-rigid regional deformation combined with multi-scale constraints is proposed. The candidate set of corresponding points is determined via the vertex’s local geometric constraints; the corresponding of partitions’ points is achieved by multi-scale geometric characteristics, the corresponding of partitions’ boundary points is achieved by differential properties and distance constraint. Experimental results show that this method improves the accuracy of the corresponding point.4. According to the algorithms proposed in this paper, design and implement a three-dimensional craniofacial point corresponding system:3DFCS. This system can achieve the three-dimensional craniofacial points corresponding relationship effectively.