# Going Deeper in Facial Expression Recognition using Deep Neural Networks

**Authors:** Ali Mollahosseini, David Chan, Mohammad H. Mahoor  
**arXiv:** [1511.04110](https://arxiv.org/abs/1511.04110)  
**Date:** 2015-11-12  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1511.04110v1/

## Key points

- Automated Facial Expression Recognition (FER) is a challenging problem in computer vision.
- Existing approaches struggle to generalize well when applied to unseen images or real-world settings.
- Many existing approaches rely on engineered features and hyperparameter tuning, limiting their effectiveness on novel data.
- The authors propose a deep neural network architecture for FER across multiple standard face datasets.
- The proposed network consists of two convolutional layers, max pooling, and four Inception layers.
- It is designed as a single component architecture that takes registered facial images as input and classifies them into one of six basic expressions or neutral expression.
- Comprehensive experiments were conducted on seven publicly available facial expression databases, showing comparable or better results than state-of-the-art methods for FER.
- The proposed architecture outperforms traditional convolutional neural networks in terms of accuracy and training time.
- This suggests that the deep neural network approach is more effective at generalizing across different datasets and capturing subtle variations in facial expressions.
- Overall, this paper presents a promising solution for automated Facial Expression Recognition using deep neural networks with improved performance compared to existing methods.

## Summary

Automated Facial Expression Recognition (FER) has long been a challenging problem in computer vision. Existing approaches have struggled to generalize well when applied to unseen images or those captured in real-world settings. Many of these approaches rely on engineered features and hyperparameter tuning, which limits their effectiveness on novel data. In this paper, the authors propose a deep neural network architecture to address the FER problem across multiple standard face datasets. The proposed network consists of two convolutional layers followed by max pooling and four Inception layers. It is designed as a single component architecture that takes registered facial images as input and classifies them into one of six basic expressions or neutral expression. The authors conducted comprehensive experiments on seven publicly available facial expression databases, including MultiPIE, MMI, CK+, DISFA, FERA, SFEW, and FER2013. The results of the proposed architecture are comparable to or better than state-of-the-art methods for FER. Notably, the proposed architecture outperforms traditional convolutional neural networks in terms of both accuracy and training time. This suggests that the deep neural network approach is more effective at generalizing across different datasets and capturing subtle variations in facial expressions. Overall, this paper presents a promising solution for automated Facial Expression Recognition using deep neural networks which shows improved performance compared to existing methods and offers potential for further advancements in this field.