Fine-Grained Facial Ethnicity Recognition Based on Dual Convolutional Autoencoders
Wing W. Y. Ng, Zixin Zhou, Ting Wang · 2021
Faces contain abundant biological and sociological information. Inter-ethnicity identification using facial images has been intensively studied, while intra-ethnicity classification has received less attention. In this paper, we propose an Ensemble of Convolutional Autoencoders (E-CAE) model to attempt to distinguish Chinese, Japanese, and Korean faces and individuals from different regions of China. To accomplish this task, CJK and RoC datasets are built and E-CAE yields a classification accuracy of 80.69% on CJK dataset and 61.81% on RoC dataset. The experimental results demonstrate that our model outperforms existing methods for fine-grained ethnicity recognition in terms of accuracy and robustness. To our knowledge, this is the first work that performs fine-grained ethnicity recognition at the scale of provinces.