Research Of Age Estimation Method Based On Facial Images

As one of the important ways of human-computer interaction,facial age estimation is widely applicable in many areas.In fact,algorithms with better performances are still solicited by varieties of applications.Along with the growth of the age,the aging process of different parts in a face varies significantly.Accordingly,they contribute differently to age estimation.However,many algorithms for age estimation based on facial images deal with all parts equally,or simply take only those most important parts into consideration.The dissertation further studies on age estimation based on facial images.Important contributions of the dissertation are as the following:1.An algorithm W-DPL for facial age estimation is put forward based on weighted dictionary pair learning model.In this algorithm,the face is firstly divided into 4×4 blocks,and the blocks are partitioned into primary zone,secondary zone and noise zone;then while the noise zone is neglected completely,the primary features and secondary features of a face are obtained by applying LBP algorithm on primary zone and secondary zone independently;subsequently,two dictionary pairs are learnt from the two kinds of features accordingly;lastly,different weights are attached to the two dictionary pairs;then facial ages are estimated by the weighted dictionary pairs.The simulation experimental results illustrate that the comprehensive performances of algorithm W-DPL is better than other analogous algorithms.2.An algorithm HS-DPL for facial age estimation is put forward based on ensemble classifiers.In algorithm HS-DPL,the face is firstly divided into 4×4 blocks;after the 6 noise blocks being removed,the features of each block is got by independently running LBP algorithm on the rest 10 blocks,then 10 sub-classifiers of dictionary pairs are learnt from the 10 features accordingly;lastly,the weights of those sub-classifiers are optimized by harmony search algorithm,and the facial ages are ultimately estimated by the ensemble classifier.The results of simulation tests on FG-NET and MORPH,two widely used open image databases in related research areas,have shown that the algorithm HS-DPL even outperforms the above algorithm W-DPL.3.A prototype system for age estimation based on facial images is designed and developed.The system includes modules of image display,model import,and age estimation.The system test results show that the system can effectively estimate age categories based on a facial image.