Human Gender Classification Based on Hand Images Using Deep Learning

EasyChair Preprint 7174

Abstract

Soft biometrics such as the gender, age, etc. can offer relevant information for person identification. The hand-based modalities are widely studied for conventional biometric recognition for various applications. However, a little research attention is grown to tackle soft biometrics using hand images. In this paper, human gender classification is addressed using the frontal and dorsal hand images. For experimentation, we have created a new hand dataset at our University, denoted as U-HD, representing sufficient posture variations at an uncontrolled environment. We have collected the sample hand images of 57 persons to incorporate more user-flexibility in posing their hands that incur additional challenges to discriminate the gender of the person. Five state-of-the-art deep neural architectures are used as the backbones, and a simple deep model is used for the human gender discrimination. The method achieves the best 90.49% accuracy on the U-HD using the Inception-V3 model.

Keyphrases: Convolutional Neural Networks, Gender Recognition, Hand biometrics, soft biometrics

@booklet{EasyChair:7174,
  author    = {Rajesh Mukherjee and Asish Bera and Debotosh Bhattacharjee and Mita Nasipuri},
  title     = {Human Gender Classification Based on Hand Images Using Deep Learning},
  howpublished = {EasyChair Preprint 7174},
  year      = {EasyChair, 2021}}