An ethnic-specific age group ranking approach to facial age estimation using raw pixel features

The age ranking approach to facial age estimation has been recently studied and has shown significant improvements in age estimation accuracy. We propose an age ranking approach which learns across three different dimensions; age groups, individual ageing patterns and ethnicity. As an extension of our earlier GroupWise age ranking model, this paper proposes a GroupWise age ranking model that is sensitive to the ethnic differences of the facial images in the dataset. With a reference image set that is arranged along these three dimensions, ranks were obtained for each age group according to the ethnicities of the facial images thus reducing the effect of ethnic confusion in age learning and subsequently improving age estimation performance. With a somewhat simplistic, but effective feature extraction and selection technique, using raw pixel values, the proposed model, when tested on FG-NET dataset, gives age estimation performance (MAE of 3.19 years) quite competitive with the state-of-the-art age estimation algorithms. The performance of the proposed model was also evaluated on a local dataset of indigenous African faces and good performance was recorded.