Personalized Digital Modeling Of Head Anatomy Guided By Facial Photograph

Personalized digital models of head anatomy have important applications in many fields such as facial surgery,electromagnetic and biomechanical simulation of the head,and body art modeling.The use of personalized head modeling guided by facial photographs becomes an alternative solution worth investigating when the tomographic imaging equipment for head imaging is not convenient to achieve.This study is based on the deformable digital anatomical atlas of Chinese population developed by the group to achieve the alignment of the 3D atlas with 2D frontal photographs and obtain personalized 3D modeling of the subject’s head and its internal tissues.The main contents of this paper can be divided into three parts as follows.(1)Facial region and feature point detection based on frontal photographs.In this study,three facial region detection algorithms are compared,including two traditional methods based on Hal features and gradient histogram features with neural network-based methods.By comparing the differences in accuracy and speed of the three methods,the directional gradient histogram-based method is finally determined as the face detection algorithm according to the operational environment requirements of this project.Based on the facial region detection,the project custom 11 face feature point detection algorithm is constructed based on the face feature point data from the open-source database,and the error of the obtained face feature points is within 4 pixels,which realizes accurate and fast automatic facial feature point detection.(2)Personalized 3D modeling of the external surface of the head.This study constructs a personalized external surface model of the head based on the facial feature points in the frontal photo,and continuously optimizes the deformation of the statistical shape model of the external surface of the head through an iterative algorithm to minimize the distance between the 2D projection of the 3D facial feature points and the feature points in the photo,so as to achieve the alignment of the 3D model with the 2D photo.Further,the facial texture in the photo is back-projected back to the 3D model surface to obtain a personalized facial texture.The experimental results show that the algorithm in this chapter not only achieves accurate 3D shape modeling of the external surface of the head for the subject,but also realistic mapping of facial textures,achieving personalized modeling effects in terms of both shape and texture.(3)Personalized 3D modeling of the complete anatomical structure of the head.Based on the 3D model of the external surface of the head,the statistical shape model of the complete anatomical structure is guided to be deformed by taking the vertices of the external surface as control points,and the internal anatomical structure is made to match the morphology of the external surface by the deformation vector interpolation algorithm,so that the personalized 3D modeling of the internal complete anatomical structure(including brain,blood vessels,nerves,bones,muscles,glands,etc.)is finally obtained.In this study,the accuracy of 3D modeling of skin and bone was evaluated using expert segmentation of CT scan images of the subject’s head as the gold standard,with an average error between 3 and 4 mm.The research in this study provides a solution for personalized 3D modeling of the complete anatomical structure of the head without relying on tomographic medical images and provides a relatively simple and fast modeling method for medical and scientific applications where internal tissue modeling error requirements are greater than 4 mm.Future research will focus on multi-angle photo modeling and improving the accuracy of internal tissue modeling.