Research On The Key Technologies Of Craniofacial Identification Based On Craniofacial Superimposition

Craniofacial Superimposition is a effective way to identify unknown skulls by the similarity comparison between the unknown skull and the real face photograph.In this thesis,the basic idea of traditional craniofacial superimposition is used to the identification of unknown skulls based on a large number of face photos which tagged with personal information.A craniofacial identity recognizer based on BP neural network is proposed to narrow the range of possible origins of unknown skulls rapidly to support criminal investigation and disaster investigation process.This thesis mainly studies the similarity measurement method between face images and 3D skull models,including the automatic extraction methods of feature points and feature lines of 3D skulls and face photos.The main research contents of this thesis include:1、Proposed a method for extracting face contour lines of improved Gradient Vector Flow(GVF)model.This thesis uses the face alignment method to calibrate the key points of two-dimensional face photos,and then constructs initial contour lines along the key points,and converges the initial contour lines with GVF,and finally extracts the accurate contour lines of 2D face photos.The experiment shows that this extraction method is fast,efficient and accurate.2、The deep learning method is applied to the computer-aided craniofacial identity.Based on the Point Net++ algorithm,the three-dimensional skull feature segmentation is realized and the mandibular region is selected,a certain conversion method is used to extract the mandibular contour line.This process automates the key technology of computer-aided craniofacial identity.3、A craniofacial identity recognizer based on BP neural network is proposed to compare similarities of the 3D skulls and face photos directly,and the number of identification indicators that meet the requirements is used as the input of the network,the output result is whether the skull and the photo may be the same person.This thesis designs a reasonable accurate network to complete the training and establish the rules of craniofacial identity recognizer.The experiment shows that the design of the craniofacial identity recognizer is reasonable,accurate and fast.4、This thesis designs and implements a prototype system of craniofacial identity.The system functions include automatic labeling of file 3D skull models and feature points of face photos,extraction of contour lines,and similarity measurement of cranial image.Experiments prove that the system is designed reasonably and can be applied to the database of real missing population,narrowing the scope of the photo collection to be identified,thereby improving the efficiency of craniofacial identity to a certain extent.