Research Of Key Techniques On Statistical Craniofacial Reconstruction

Craniofacial reconstruction, which is used to infer the original appearance of the unknown skull based on the inherent law between skull and appearance, has significant application value in the fields such as archeology, forensic medicine and medical plastic. Statistical craniofacial reconstruction method is more scientific and objective than traditional reconstruction methods, and it has recently become a research focus in craniofacial morphology and reconstruction. With the support of "The researches on Craniofacial Morphology and Craniofacial Reconstruction"(key project of National Nature Science Foundation), comprehensively applying image and graphics processing technology, the theory of multivariate statistical analysis, and forensic anthropological knowledge, this dissertation, which aims to explore the statistical laws of the morphological relationship between skull and face, deeply researched key techniques on statistical craniofacial reconstruction, including the construction of3D craniofacial model database, the location of feature points, the establishment of physiological consistent correspondence for craniofacial models, the statistical modeling on craniofacial morphological relationship and its application in craniofacial reconstruction. The main contributions of this dissertation are as following:1. Surface reconstruction of craniofacial models based on CT images is researched. Aiming at surface reconstruction of craniofacial models based on3D CT images of living samples, the improved GVF-Snake model whose parameters are optimized by force field of Gradient Vector Flow(GVF), which improves the accuracy of the craniofacial outer contour extraction, is researched. A contour-based craniofacial model reconstruction algorithm in which the improved GVF-Snake model and ray method are combined is put forward. Simulating the process of human eye observing3D objects, a new craniofacial model reconstruction algorithm based on Breadth First Search(BFS) algorithm of graph and multi-view visibility detection is proposed, by which automatic and efficient surface reconstruction of craniofacial models is basically realized.2. The construction of3D craniofacial model database including skulls and faces is also researched. A coordinates correction approach based on Ordinary Least Square(OLS) linear regression is proposed, by which Frankfurt Coordinate system of craniofacial model is established accurately. A3D mesh repair algorithm for craniofacial model based on the physiological symmetry and improved layer-by-layer pulsive hole-filling algorithm for repairing defective craniofacial model based on normal constraint of adjacent triangular faces, which enable consistent curvature between the filled meshes and its surrounding, are presented.3D model cutting algorithms based on VTK and mesh simplification algorithm based on level-set method are present, and efficient cutting and simplifying of craniofacial model are realized. A classified3D craniofacial model database is established by adopting anthropological classification method.3. The location of feature points is researched. Two sets of feature points of skull and face are defined according to craniofacial physiological structural characteristics, and a new feature points location method for skull and face based on Multi-scale Geometric Features Similarity Distance(MGFSD), employing classified feature points template and hierarchical screening process, is proposed. At first, the Preliminary Candidate Set(PCS) of feature points is established under the guidance of classified feature points template. Next, the Effective Candidate Set(ECS) is obtained from PCS by hierarchical screening process based on similarities of normals and local statistical characteristics of corresponding feature points. Finally, according to MGFSD based on volume integral invariants, the optimal candidate in ECS can be determined as feature points. It improves the location accuracy of feature points, and automatic location of the key feature points in craniofacial model is effectively realized.4. The establishment of physiological consistent correspondence of3D craniofacial models is researched. The ICP-based correspondence algorithm for craniofacial models is improved by introducing the weighted distance constraint of multiple features. A hierarchical3D facial correspondence method based on regional deformation and multi-constraint and a3D cranial correspondence method based on global TPS deformation and multi-constraint are proposed by introducing local relative position geometric constraint. Experimental results demonstrate our methods improve the correct rates of correspondence substantially and can establish physiological correspondence with similar local shape between craniofacial models effectively.5. Statistical modeling on craniofacial morphological relationship and its application in craniofacial reconstruction is deeply researched. Firstly, the construction of the combined craniofacial shapes statistical model based on PCA is researched and the combined model is applied into craniofacial reconstruction, whose defects are pointed out according to the experimental results. Secondly, introducing partial least squares regression(PLSR) into craniofacial reconstruction, the statistical regression model of local craniofacial morphological correlation is first put forward, and a new statistical craniofacial reconstruction method based on this PLSR model is proposed. Thirdly, an evaluation algorithm based on the similarity of corresponding point sets is present in this dissertation and applied in the evaluations of the two craniofacial reconstruction methods. Experimental result demonstrates the new statistical craniofacial reconstruction method based on PLSR can improve the reconstruction accuracy substantially.