Ethnicity Classification Based on Facial Images using Deep Learning Approach

Race and ethnicity are terminologies used to describe and categorize humans into groups based on biological and sociological criteria. One of these criteria is the physical appearance such as facial traits which are explicitly represented by a person’s facial structure. The field of computer science has mostly been concerned with the automatic detection of human ethnicity using computer vision-based techniques, where it can be challenging due to the ambiguity and complexity on how an ethnic class can be implicitly inferred from the facial traits in terms of quantitative and conceptual models. The current techniques for ethnicity recognition in the field of computer vision are based on encoded facial feature descriptors or Convolutional Neural Network (CNN) based feature extractors. However, deep learning techniques developed for image-based classification can provide a better end to end solution for ethnicity recognition. This paper is a first attempt to utilize a deep learning-based technique called vision transformer to recognize the ethnicity of a person using real world facial images. The implementation of Multi-Axis Vision Transformer achieves 77.2% classification accuracy for the ethnic groups of Asian, Black, Indian, Latino Hispanic, Middle Eastern, and White.

@article{Kalkatawi2024,
  title     = {Ethnicity Classification Based on Facial Images using Deep Learning Approach},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {2},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Abdul-aziz Kalkatawi and Usman Saeed},
  doi       = {10.14569/IJACSA.2024.0150223},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150223}
}