SAFER: Situation Aware Facial Emotion Recognition

TL;DR

SAFER is a deep learning-based facial emotion recognition system that incorporates contextual information and adapts to unseen expressions, achieving high accuracy and addressing dataset limitations for real-world applications.

Contribution

The paper introduces SAFER, a novel emotion recognition system that combines facial features with context and proposes a new dataset to overcome existing limitations.

Findings

Achieves 91.4% accuracy on CAER-S dataset

Enhances recognition in open-world settings

Addresses challenges posed by face masks during COVID-19

Abstract

In this paper, we present SAFER, a novel system for emotion recognition from facial expressions. It employs state-of-the-art deep learning techniques to extract various features from facial images and incorporates contextual information, such as background and location type, to enhance its performance. The system has been designed to operate in an open-world setting, meaning it can adapt to unseen and varied facial expressions, making it suitable for real-world applications. An extensive evaluation of SAFER against existing works in the field demonstrates improved performance, achieving an accuracy of 91.4% on the CAER-S dataset. Additionally, the study investigates the effect of novelty such as face masks during the Covid-19 pandemic on facial emotion recognition and critically examines the limitations of mainstream facial expressions datasets. To address these limitations, a novel dataset for facial emotion recognition is proposed. The proposed dataset and the system are expected to be useful for various applications such as human-computer interaction, security, and surveillance.