Research On Craniofacial Point Correspondence Algorithm And Reconstruction Algorithm Based On Statistical Regression

Craniofacial reconstruction refers to the technique of recovering the appearance of the deceased according to the acquired morphological relationship between skulls and faces after identifying the sex,race,and age of an unknown skull.It has high research value in archeology,forensic medicine and other fields.With the increase of craniofacial collected data,the craniofacial reconstruction method based on statistical model has gradually became mainstream.The key to the statistical craniofacial reconstruction method is to accurately establish the physiological points correspondence between craniofacial models,and the statistical models of skulls to faces.In view of this two key problems,the work is made as follow: 1)Craniofacial data preprocessing: Three dimensional reconstruction of living craniofacial CT data was performed,and the models was denoised,hole-filled,and data normalization.Based on the analysis of the MPEG standard,78 craniofacial feature points were defined and marked.2)Craniofacial model registration and point correspondence algorithm.In the coarse registration phase,the control points required for the TPS were randomly generated,and the non-rigid registration of the skull and the facial skin model was achieved with the ICP algorithm;A linear combination based on iterative model was proposed to achieve the accurate registration of the craniofacial model.Combined with the multiple constra ints of the integral invariants,the physiological point correspondence between skulls or faces was established.The experimental results show that on the basis of the registration method in this paper,the accuracy of the point correspondence is higher.3)Craniofacial reconstruction based on Least Squares Support Vector Regression.Firstly,statistical models of skulls and faces were established by Principal Component Analysis(PCA).The target model was trained by using LSSVR in the shape parameter space.Then the unknown skull is placed in the shape parameter space,and reconstructive face is closer to the real face.4)A craniofacial reconstruction system based on LSSVR was designed.The functions of craniofacial sample data registration,establishment of point correspondence,establishment of statistical model,craniofacial reconstruction,and error estimation were realized.