Studies On Prediction Of Three-dimensional Facial Age

Human facial morphology is one of the important biological characteristics of human,which plays an important role in individual recognition and social activities.Facial features will show visible signs of aging with age.In the field of forensic science,this change will bring great difficulties to the development of forensic medicine,video image investigation and other works,which will affect the efficiency of solving cases.Therefore,the research of age estimation and reconstruction is of great practical significance in criminal investigation,social security and other fields.However,the accuracy of age estimation is not high in previous studies and the effect of aging simulation is not ideal.Two-dimensional images based on foreign public face databases were used in most studies,but few researches were carried out to apply threedimensional images of Chinese population.Consequently,in order to provide powerful technical support for elaboration of facial images,it is urgent to study the aging patterns of various regional and ethnic groups in China.In this paper,we applied the partial least square regression(PLSR)to analyze the 3D facial images and rebuild a age estimation model in Chinese populations.Two main research aspects of this paper were addressed as followed:(1)Age estimation and reconstruction via three-dimensional facial images in Xinjiang Uygur males.Firstly,we calculated the average face of different age groups.Average faces exhibit the eye corners droop,the cheek depressed,the nasolabial sulcus become deeper and other characteristics of aging with age.And then a PLSR regression model of this population was constructed to implement age estimation.The pearson correlation coefficient(PCC)between predict ages and chronological ages was 0.71,which indicated significant correlation degree.The mean absolute deviation(MAD)of age estimation was 6.37 years.Among these results,31-40 years group gave the best prediction accuracy(MAD 4.27).Heat maps based on the regression coefficient showed that the influencing degree of different aging regions exerts effect on age models,where these regions were consistent with aging parts identified by human eyes visually.Finally,based on the PLSR model,the images of five individuals were reconstructed to form younger and older images respectively.The facial morphologies were visualized to synthesize,and our aging model had a better effect.(2)In-depth study of facial aging patterns among different regional and ethnic groups in China.On the basis of single group study,through calculating the MAD value of various groups in different PLSR components,we took 4 PLSR components as the principal thing and 5 PLSR components as the supplement to study aging patterns of multiethnic groups.Multidimensional Scaling analysis(MDS)based on regression coefficient indicated that aging patterns of different district ethnic groups had a double effect of geographical and environmental factors.Heat maps based on the regression coefficient of each group showed that the ethnic groups with closer distance were more similar in their aging characteristics.Finally,we reconstructed the aging faces of five age groups in different populations by establishing aging vectors,whose aging faces were highly correlated with the MDS results.Not only did reconstructive aging faces have strong personalized characteristics,but it could reflect aging process veritably and had a better aging simulation individually.So it is of great value for meticulous depiction of suspect portrait.