Gender Classification Based on Body Images
Bingze Dai, Dequan Yang · 2021 China Automation Congress (CAC) · 2021
The gender classification for human being is one of the interesting problems in computer vision and plays an important role in various daily applications. Especially during the pandemic, people are wearing masking blocking their faces, extracting body information becomes much more important. In this paper, we investigate the problem of predicting gender from 2D human body images using various deep neural network frameworks. Our proposed method does not acquire other biometrics feature extraction and calculation which is an end- to-end method. This work demonstrates estimating soft biometric characteristics such as gender with mere body images without faces can also achieve high accuracy. We also proposed a pipeline to clean human body images to leave only body information and created a labeled 2D image dataset containing only body information despites the face and background information. Several different modified deep neural networks are tested and compared on this dataset. The best model is modified from Resnet50 and it provides relative high accuracy for gender classification using pure human body images.