# Multi-Expert Gender Classification on Age Group by Integrating Deep Neural Networks

**Authors:** Jun Beom Kho  
**arXiv:** [1809.01990](https://arxiv.org/abs/1809.01990)  
**Date:** 2018-09-06  
**Summary page:** https://www.summarizepaper.com/en/paper/1809.01990v2/

## Key points

- Accurate gender classification in facial recognition technology is crucial but can be impacted by variations in facial age.
- Changes in facial shape and skin texture occur as individuals age, requiring a re-evaluation of existing gender classification systems to incorporate information about facial age.
- The Multi-expert Gender Classification on Age Group (MGA) approach is proposed to address this issue.
- MGA utilizes an end-to-end multi-task learning framework for age estimation and gender classification tasks.
- It employs two deep neural networks: Convolutional Appearance Network (CAN) for facial appearance features and Deep Geometry Network (DGN) for facial geometric features.
- Model integration strategy and fine-tuning are used to enhance accuracy of both age estimation and gender classification.
- Facial images are categorized into young, adult, or elder based on estimated age, followed by gender predictions using an average fusion strategy from three gender classification experts within each age group.
- Experimental evaluations show that the MGA system outperforms state-of-the-art approaches while requiring smaller computational resources.

## Summary

In the field of facial recognition technology, accurate gender classification is crucial but can be significantly impacted by variations in facial age. As individuals age, changes in facial shape and skin texture occur, requiring a re-evaluation of existing gender classification systems to incorporate information about facial age. To address this issue, a novel approach called Multi-expert Gender Classification on Age Group (MGA) is proposed in this paper. The MGA system utilizes an end-to-end multi-task learning framework that simultaneously performs age estimation and gender classification tasks. It employs two deep neural networks: the Convolutional Appearance Network (CAN) for extracting facial appearance features and the Deep Geometry Network (DGN) for capturing facial geometric features. These networks are integrated using a model integration strategy and fine-tuned to enhance the accuracy of both age estimation and gender classification. One key aspect of the MGA system is its categorization of facial images into one of three age groups: young, adult, or elder based on estimated age. Subsequently, the system makes gender predictions using an average fusion strategy from three gender classification experts trained to recognize specific gender characteristics within each age group. Extensive experimental evaluations conducted on challenging databases demonstrate that the proposed MGA system outperforms several state-of-the-art approaches while requiring smaller computational resources. This research represents a significant advancement in improving gender classification accuracy in facial recognition systems by effectively integrating deep neural networks and considering the impact of facial age variations on classification performance.