Research On Regional Based Craniofacial Reconstruction Technology And Realistic Processing Methods

Craniofacial reconstruction technology is a method to reproduce the facial appearance of an unidentified skull. This technology is based on one of the theories of physical anthropology that the appearance of face is based on the shape of skull. It is mainly used in anthropology, archeology, forensic science, craniofacial plastic surgery and other fields. Computer-aided craniofacial reconstruction technology, as an important part of informatics craniofacial morphology research, is more objectiv and efficient than the traditional manual reconstruction methods, and become a hot area of information technology research recently at home and abroad. This thesis is under the support of the Key Project of National Natural Science Fundation "Research of craniofacial morphology and reconstruction (No.60736008)" and focus on several disadvantages of current methods of computer-aided craniofacial reconstruction methods, integrates the knowledge of physical anthropology, computer graphics and image processing techniques and statistical analysis methods to improve the framework of automatic computer-aided craniofacial reconstruction system. This thesis introduces the construction process of craniofacial database of the Mongoloid and focuses on the research of regional craniofacial reconstruction method based on partial least squares regression model, fast and reusable hair modeling method and facial texture generating method based on texture synthesis. Based on those methods, high realistic Mongolia craniofacial reconstruction framework is implemented. The main work and innovations of this thesis are listed as follows:1. Craniofacial database construction and reconstruction of single layer closed three-dimensional models of face and skull. Focus on the lack of basic data for Mongoloid craniofacial reconstruction, this thesis designs and improves the norms and methods in craniofacial medical image data acquisition and builds a more complete and objective data foundation for craniofacial reconstruction research; Presented a single closed three-dimensional model reconstruction methods based on curve evolution to implement non-redundant and efficient craniofacial surface reconstruction.2.3D craniofacial model decompose and merging methods. Focus on the insufficient and low realistic results of craniofacial reconstruction methods based on whole craniofacial models, the definition of feature point set and a method for automatic feature points marking by depth image segmentation and gray integral projection is introduced. A craniofacial three-dimensional model decompose method based on feature points constrains is proposed to decompose craniofacial models into characteristic parts; A craniofacial parts model merging method which is based on local metamorphosis fusion is proposed to implement the smooth merging of reconstructed facial parts.3. Regional craniofacial reconstruction method based on partial least squares regression. For the poor detail of reconstructed face from current craniofacial reconstruction methods, a regional craniofacial reconstruction method based on partial least squares regression model is proposed. Firstly, a3D craniofacial point correspondence method based on geodesic distance consistency constraint is introduced to establish the point correspondence between parts from different craniofacial models. Secondly, a regional craniofacial reconstruction method based on partial least squares regression model is presented. The results of craniofacial reconstruction by this method is compared to the results of PCA-based statistical craniofacial reconstruction methods, which indicate that unidentified faces reconstructed by the method introduced by this thesis is more accurate and has more detail than results reconstructed by PCA-based methods.4. Reusable fast hair modeling method. To improve the realistic of reconstructed facial models, a new hair styling method combined with multiple hairstyle constraints is proposed. This method improves the efficiency and flexibility of interactive hair styling, and allows users to create more complex hair style. A non-interactive hair styling method based on the hair prototype is presented for model reuse requirements in masses of hair styling, which requires that the prototype model should be constructed using multiple hairstyle constraints algorithm at first. Result shows that the performance of the masses of hair styling is raised by this method.5. Craniofacial skin texture generation method. For the lack of facial skin texture of the reconstructed facial models, a texture generating method is proposed. Firstly, for most area of the reconstructed faces, facial skin texture generation method based on surface texture synthesis is proposed. For the eye area, a texture mapping method is proposed to generate the texture. To the nose and mouth area, surface texture synthesis is used to generate texture. Experiment shows that the sense of reality of the reconstrcted face is enhanced by this method.This thesis presented an effective framework for computer-aided craniofacial reconstruction. A regional craniofacial reconstruction method and a realistic modeling method are introduced. Those methods solved the problem of low accuracy and poor detail performance of traditional facial reconstruction methods. The computer-aided craniofacial reconstruction framework is improved. The applications of computer-aided craniofacial reconstruction are greatly promoted in archeology, forensic and other fields.