Facial Emotion Recognition in the Deep Learning Era: A Systematic Review of Methods, Datasets and Open Challenges, with a Proposed Context-Aware Hybrid Framework

Facial emotion recognition (FER) has moved from a laboratory curiosity to a core perception capability for healthcaremonitoring, adaptive learning, human–robot interaction, driver safety and affective computing. This article presents a structurednarrative review of 30 curated sources published between 2022 and 2026 that together span ten survey articles, two comparative or framework studies and eighteen technique-oriented contributions. We organise the literature along a taxonomy that covers handcrafted pipelines, convolutional networks, attention-enhanced networks, CNN–recurrent hybrids, Vision Transformers, graph- and action-unitbased models, self-supervised and generative learning, multimodal fusion and human-guided explainable systems. The synthesis shows a consistent drift from texture descriptors to deep convolutional backbones and, more recently, to attention and Transformer architectures, yet it also reveals persistent gaps: ambiguous and noisy labels, class imbalance, weak cross-dataset generalisation, limited temporal and contextual reasoning, scarce interpretability, and unresolved fairness and privacy concerns. Motivated by these gaps, we propose a Context-Aware Hybrid FER framework (CAH-FER) that fuses a convolutional stem, a local-attention Transformer and an action-unit graph branch, adds temporal encoding, self-supervised pre-training and uncertainty-aware learning, and ships with an explainability module. We specify a reproducible evaluation protocol. No empirical results are claimed for the proposed framework; the protocol is designed so that its hypotheses can be confirmed or refuted in follow-up experimental work.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24359272
Primary Topic
Emotion and Mood Recognition
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Facial Emotion Recognition in the Deep Learning Era: A Systematic Review of Methods, Datasets and Open Challenges, with a Proposed Context-Aware Hybrid Framework

Vanishri Jantagal, Sanjeev Kulkarni
Zenodo (CERN European Organization for Nuclear Research)
Emotion and Mood Recognition
article

Facial Emotion Recognition in the Deep Learning Era: A Systematic Review of Methods, Datasets and Open Challenges, with a Proposed Context-Aware Hybrid Framework

Vanishri Jantagal, Sanjeev Kulkarni
article en

Abstract

Facial emotion recognition (FER) has moved from a laboratory curiosity to a core perception capability for healthcaremonitoring, adaptive learning, human–robot interaction, driver safety and affective computing. This article presents a structurednarrative review of 30 curated sources published between 2022 and 2026 that together span ten survey articles, two comparative or framework studies and eighteen technique-oriented contributions. We organise the literature along a taxonomy that covers handcrafted pipelines, convolutional networks, attention-enhanced networks, CNN–recurrent hybrids, Vision Transformers, graph- and action-unitbased models, self-supervised and generative learning, multimodal fusion and human-guided explainable systems. The synthesis shows a consistent drift from texture descriptors to deep convolutional backbones and, more recently, to attention and Transformer architectures, yet it also reveals persistent gaps: ambiguous and noisy labels, class imbalance, weak cross-dataset generalisation, limited temporal and contextual reasoning, scarce interpretability, and unresolved fairness and privacy concerns. Motivated by these gaps, we propose a Context-Aware Hybrid FER framework (CAH-FER) that fuses a convolutional stem, a local-attention Transformer and an action-unit graph branch, adds temporal encoding, self-supervised pre-training and uncertainty-aware learning, and ships with an explainability module. We specify a reproducible evaluation protocol. No empirical results are claimed for the proposed framework; the protocol is designed so that its hypotheses can be confirmed or refuted in follow-up experimental work.

Zenodo (CERN European Organization for Nuclear Research)
Srinivas University (IN)
Emotion and Mood Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon