Research On The Method Of Sex Identification Based On 3D Skull Model

The digital skull has the possession of research significance and practical value in the criminal,facial reconstruction and medical plastic fields.In the process of traditional sexual identification,well-experienced professional knowledge is absolutely necessary,which is not easy to avoid subjective errors for sexual identification rate.With the development of sexual identification technology,the digital skull information has found an increasingly wide utilization in sexual identification,however,the traditional methods still need spending a lot of manpower in achieving the significance features of the 3D skull models.In view of the above-mentioned difficult questions,we put forward two methods,one is the 3D skull morphology quantification representation and nonlinear method,which reduces the workload of the researchers and improves the classification performance,and another is the automatic sex identification based on the multi-angle skull images,which realizes the sexual identification and alleviates the problem of the manpower.According to the order of this thesis,the main points of the thesis are researched as listed below.First,the standardization method of the 3D skull models is described in detail,it also explores into the significance feature points of 3D skull models,the calibration technology and the theory of measurable characteristics,which lays a theoretical foundation for acquiring the characteristics of 3D skulIs.Second,a method for quantitative representation of skull morphology and nonlinear sexual identification is presented.In this thesis the calibration feature points are achieved by self-developed measurement system in lab,then the specific measurable indicators and non-measurable indicators are achieved,and the non-measurable indicators are quantified during the experiments.We put forward to integrate several reduction methods and multiple classifiers to realize the sexual identification,the gender recognition rate increased by about 2%compared to the linear classification method.Third,research on the sexual identification based on the multi-angle skull images.The multi-angle skull images are obtained based on the normalized skull models,the convolution neural network algorithm is used to get the mean probability of each image in the skull samples belonging to the male and female,the least square method is employed in achieving the weight of each image in each sample,and then the method is achieved based on the multi-angle skull images.Finally,we design and implement the sexual identification prototype system.It demonstrates the feasibility of the two proposed methods,tests and verifies the application on the unknown skull.