Facial Age Estimation Based On Multivariate Multiple Regression

Facial age estimation is an important problem in the field of computer image processing.Because of the difficulty of data collection,one of the most challenges of facial age estimation model says there is not sufficient training data.Label distribution learning is an effective method to address this problem,where its motivation is that facial aging information on adjacent ages can be introduced to enhance the age estimation model due to the fact that human faces change slowly on adjacent ages.Given a certain age to learn,label distribution learning converts the learning target from a continuous value to an age label distribution which is generated according to the description degree of the neighboring ages to the given age.Despite of the successful application of label distribution learning on facial age estimation,the existed methods have some obvious drawbacks.The method based on the maximum entropy builds separated model for each age and has a strong assumption about the data distribution,and the neural network based method has the problem of over-fitting.Based on the existing research,this paper proposed a method facial age estimation based on multivariate multiple regression.The main innovative work is as follows.(1)The facial age estimation problem was translated into a multivariate multiple regression analysis.The multivariate multiple regression can build an integrated model for all ages,and predict the corresponding age with input facial features.(2)Multivariate partial least squares regression was used to build model.The proposed method has no assumption about the data distribution,and can deal with the potential multiple collinear problems between variables.It can build more effective models for all ages and save modeling time.In this thesis,the FG-NET and MORPH-II facial image database experiments are carried out to verify the effectiveness of proposed method.The experimental results show that the proposed method can achieve better performance and shorter training time.