Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition
Open accessThe DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index; 7 of the 17 sources its claims cite were checked for support.
Abstract
This paper proposes two multimodal fusion methods between brain and peripheral signals for emotion recognition. The input signals are electroencephalogram and facial expression. The stimuli are based on a subset of movie clips that correspond to four specific areas of valance-arousal emotional space (happiness, neutral, sadness, and fear). For facial expression detection, four basic emotion states (happiness, neutral, sadness, and fear) are detected by a neural network classifier. For EEG detection, four basic emotion states and three emotion intensity levels (strong, ordinary, and weak) are detected by two support vector machines (SVM) classifiers, respectively. Emotion recognition is based on two decision-level fusion methods of both EEG and facial expression detections by using a sum rule or a production rule. Twenty healthy subjects attended two experiments. The results show that the accuracies of two multimodal fusion detections are 81.25% and 82.75%, respectively, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%). The combination of facial expressions and EEG information for emotion recognition compensates for their defects as single information sources.
Overview
CoChat Summary
TL;DR
This paper proposes two decision-level fusion methods combining facial expressions and EEG for four-class emotion recognition (happiness, neutral, sadness, and fear). Across twenty healthy subjects, fusion via a sum rule and a production rule achieved 81.25% and 82.75% accuracy, higher than facial expression (74.38%) or EEG (66.88%) alone.
Key findings
- The two multimodal fusion detections achieved accuracies of 81.25% (sum rule) and 82.75% (production rule), both higher than facial expression detection (74.38%) or EEG detection (66.88%).
- Seventeen of 20 subjects achieved the highest accuracies using the fusion methods.
- Paired t-tests showed higher accuracies for both fusion methods than for facial expression detection (p = 0.03 for each) or EEG detection (p < 0.01 for each); facial expression versus EEG detection was not significantly different (p = 0.08), and the first versus second fusion methods were not significantly different (p = 0.56).
- Low accuracies were obtained for subjects 5, 19, and 20, which the authors attribute to less expressive facial expressions or the approach being less sensitive to them.
Methods
- Twenty healthy 19- to 33-year-old subjects from the local research unit attended two experiments, offline and online. Experiment 1 collected 40 trials per subject using movie clips for happiness, neutral, sadness, and fear; each trial collected 100 face images and 200 groups of EEG signals, with valence/arousal ratings via the Self-Assessment Manikin. Experiment 2 had 40 trials per subject and used online detectors with feedback.
- EEG was recorded with a Mindwave Mobile device from the “Fz” electrode referenced to the right mastoid, with impedances below 5 kΩ; facial expressions were captured with a Logitech camera (25 FPS, 800 × 600). EEG was bandpass filtered into eight bands, PSD features were computed using STFT with a 1-s non-overlapping Hanning window, and two linear SVM classifiers detected emotion state and intensity. Facial expression detection used AdaBoost/Haar face detection, resizing to 48 × 48, PCA to 169 dimensions, and a feedforward neural network.
- Decision-level fusion combined classifier outputs using either a sum rule or production rules.
Limitations
- The paper states that the image data set obtained was very limited and that EEG signals used for analysis were recorded from only one electrode; future work is proposed to collect more image data from more subjects, use a more complicated model, and use an EEG device with more electrodes.
- Evident from the text: the sample was twenty healthy subjects aged 19–33 from the local research unit, and the single Fz EEG electrode limits spatial coverage.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



Verification
Claims the paper makes
4 Valid · 3 Not valid · 6 Not checked · 26 Own results and interpretations
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Most commercial biometric systems do not provide access to the feature sets nor the raw data which they use in their products [22].”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 5. Discussions
What the sources say
- Could not verifycites [22] Ross A. Jain A. K. Multimodal biometrics: an overview Proceedings of the 12th European Signal Processing Conference 2004 IEEE 1221 1224 sourceExperimental — accuracy not yet measured
The Discussion mentions commercial EEG systems but says nothing about whether they provide access to raw data or feature sets. The Experiment Setup describes access to this study’s dataset, not the access policies of most commercial biometric systems. Searches for commercial-system data and feature access found no passage addressing the claim; page numbers were not available in our copy. (LLM, conf=0.74)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The notion that combining brain and peripheral physiological signals will result in a more accurate emotion recognition compared to using these variables on their own seems very sensible and has frequently been suggested in the literature as a potential way to improve emotion recognition [16].”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 5. Discussions
What the sources say
- Partly supportedcites [16] Hogervorst M. A. Brouwer A.-M. van Erp J. B. F. Combining and comparing EEG, peripheral physiology and eye-related measures for the assessment of mental workload Frontiers in Neuroscience 2014 8, article 322 10.3389/fnins.2014.00322 2-s2.0-84926647412 PMC4196537 25352774 sourceExperimental — accuracy not yet measured
The Combination of Information section says that combining physiological variables is frequently suggested as a way to improve mental-state assessment, but it does not establish that it will improve emotion recognition. The Combining Variables–Fusion Techniques section reports a small gain in one emotion-classification study and no strong advantage in another. The source therefore supports the proposal as a possibility, not the claim of a reliable improvement for emotion recognition. (LLM, conf=0.85)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Superior performance was obtained compared to the results in the state-of-the-art results [3, 11, 32].”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 5. Discussions
What the sources say
- Not matchedcites [3] Mase K. Recognition of facial expression from optical flow IEICE Transactions 1991 74 10 3474 3483
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [11] Lang P. J. Bradley M. M. Cuthbert B. N. International affective picture system (IAPS): affective ratings of pictures and instruction manual 2008 A-8
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites [32] Zheng W.-L. Lu B.-L. A multimodal approach to estimating vigilance using EEG and forehead EOG Journal of Neural Engineering 2017 14 2 026017 10.1088/1741-2552/aa5a98 2-s2.0-85015796263 28102833 source
The citation is attached to 'the state-of-the-art results [3, 11, 32]', i.e. it identifies Zheng & Lu (2017) as one of the published baseline results in EEG/EOG vigilance estimation against which the citing authors compare. Our copy is exactly that: a peer-reviewed multimodal vigilance-estimation study reporting quantitative performance (Conclusion, Section 5: single forehead EOG 0.78/0.12 and EEG 0.70/0.13 COR/RMSE, rising to 0.85/0.09 with modality fusion plus temporal dependency; Section 3.3 gives SVR/CCRF/CCNF 0.83/0.84/0.85). Note that the source itself claims only 'comparable performance with the conventional methods on critical brain areas' (Section 4, Discussion) and superiority over its own single modalities, not over other published work — but the 'superior performance' assertion belongs to the citing paper, not to this source, so it is not judged here. (LLM, conf=0.82)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results show that the accuracies of two multimodal fusion detections are 81.25% and 82.75%, respectively, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%).”
- Also stated in: § 4. Results · plain
“It shows that the accuracy of the first fusion detection using a sum rule is 81.25% and the accuracy of the second fusion detection using a production rule is 82.75%, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%).”
- Also stated in: § 5. Discussions · plain
“The results show that the accuracies of two information fusion detections are 81.25% and 82.75%, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%).”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > Abstract
plainn = 20 ·- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Also stated in: § 4. Results · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results show that the accuracies of two multimodal fusion detections are 81.25% and 82.75%, respectively, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%).”
- Also stated in: § 4. Results · plain
“It shows that the accuracy of the first fusion detection using a sum rule is 81.25% and the accuracy of the second fusion detection using a production rule is 82.75%, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%).”
- Also stated in: § 5. Discussions · plain
“The results show that the accuracies of two information fusion detections are 81.25% and 82.75%, which are both higher than that of facial expression (74.38%) or EEG detection (66.88%).”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > Abstract
plainn = 20- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Also stated in: § 4. Results · plain
20 more claims
Create a free account to see all 39 claims and every source verdict.
- Probably not a substantial claim. 9
- It may state two separate claims as one. 6
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Using fusion techniques at the feature and decision levels, Zheng et al. [12] showed, for example, that combining EEG and eye movement data could improve emotion recognition performance.”
From Multimodal Emotion Recognition: Emotion Classification Through the Integration of EEG and Facial Expressions (2025) · cites this paper as [15]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Huang et al. achieved a prediction accuracy of 74.38% using only EEG data and 66.38% using only facial expression data, both of which are unimodal methods, for the four basic emotional states in the their custom dataset. However, by implementing two different data fusion techniques, they improved these accuracy rates to 81.25% and 82.75%.”
From Multimodal Emotion Recognition: Emotion Classification Through the Integration of EEG and Facial Expressions (2025) · cites this paper as [15]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Studies by Ma et al. [14] and Huang et al. [15] used EEG with other physiological signals and custom datasets, achieving accuracies of 92.87% and 82.75%, respectively.”
From Multimodal Emotion Recognition: Emotion Classification Through the Integration of EEG and Facial Expressions (2025) · cites this paper as [15]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 1. Introduction
“All of the studies have shown that the performances of emotion recognition systems can be improved by employing multimodal information fusion.”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 1. Introduction
“the use of self-assessment labels for arousal assessment yielded accuracies of 55%, 53%, and 54% for EEG, physiological, and fused features, respectively.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 1. Introduction
“Picard and Daily at MIT Media Laboratory developed pattern recognition algorithms that attained 78.4% classification accuracy for three categories of emotion states using the peripheral signals of galvanic skin resistance, blood pressure, respiration, and skin temperature [4].”
§ Fusion of Facial Expressions and EEG for Multimodal Emotion Recognition > 5. Discussions
“In fact, the authors of [3] reported an average accuracy of 78.4% by using optical flow to determine the main direction of movement of the muscles.”
Citations
Cites
Measuring emotion: The self-assessment manikin and the semantic differential
Bradley, M. M., & Lang, P. J. (1994) · Journal of Behavior Therapy and Experimental Psychiatry · doi:10.1016/0005-7916(94)90063-9 · Cited by 9,538 · cited 1×
“Valence and arousal ratings were obtained using the Self-Assessment Manikin (SAM) [15].”
MethodA review of classification algorithms for EEG-based brain–computer interfaces
Lotte, F., Congedo, M., Lécuyer, A., Lamarche, F., & Arnaldi, B. (2007) · Journal of Neural Engineering · doi:10.1088/1741-2560/4/2/r01 · Cited by 2,665
Information fusion in biometrics
Ross, A., & Jain, A. (2003) · Pattern Recognition Letters · doi:10.1016/s0167-8655(03)00079-5 · Cited by 1,107
Incremental support vector learning for ordinal regression
Gu, B., Sheng, V. S., Tay, K. Y., Romano, W., & Li, S. (2015) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2014.2342533 · Cited by 707
A Robust Regularization Path Algorithm for $\nu $ -Support Vector Classification
Gu, B., & Sheng, V. S. (2017) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2016.2527796 · Cited by 449
A multimodal approach to estimating vigilance using EEG and forehead EOG
Zheng, W.-L., & Lu, B.-L. (2017) · Journal of Neural Engineering · doi:10.1088/1741-2552/aa5a98 · Cited by 424 · cited 1×
“Superior performance was obtained compared to the results in the state-of-the-art results [3, 11, 32].”
ResultA Hybrid BCI System Combining P300 and SSVEP and Its Application to Wheelchair Control
Li, Y., Pan, J., Wang, F., & Yu, Z. (2013) · IEEE Transactions on Biomedical Engineering · doi:10.1109/tbme.2013.2270283 · Cited by 421
Self-citationShort-term emotion assessment in a recall paradigm
Chanel, G., Kierkels, J. J. M., Soleymani, M., & Pun, T. (2009) · International Journal of Human-Computer Studies · doi:10.1016/j.ijhcs.2009.03.005 · Cited by 367
EEG-based neonatal seizure detection with Support Vector Machines
Temko, A., Thomas, E., Marnane, W., Lightbody, G., & Boylan, G. (2011) · Clinical Neurophysiology · doi:10.1016/j.clinph.2010.06.034 · Cited by 339 · cited 1×
“It should be noted that SVM classifier was widely used in the EEG-based brain computer interface in practice [26–29].”
MethodHogervorst, M. A., Brouwer, A.-M., & van Erp, J. B. F. (2014) · Frontiers in Neuroscience · doi:10.3389/fnins.2014.00322 · Cited by 292 · cited 1×
“The notion that combining brain and peripheral physiological signals will result in a more accurate emotion recognition compared to using these variables on their own seems very sensible and has frequently been suggested in the literature as a potential way to improve emotion recognition [16].”
BackgroundLeeb, R., Sagha, H., Chavarriaga, R., & Millán, J. D. R. (2011) · Journal of Neural Engineering · doi:10.1088/1741-2560/8/2/025011 · Cited by 242
Evaluating a Computational Model of Emotion
Gratch, J., & Marsella, S. (2005) · Autonomous Agents and Multi-Agent Systems · doi:10.1007/s10458-005-1081-1 · Cited by 161
Facial Expression Recognition Using Constructive Feedforward Neural Networks
Ma, L., & Khorasani, K. (2004) · IEEE Transactions on Systems, Man and Cybernetics, Part B (Cybernetics) · doi:10.1109/tsmcb.2004.825930 · Cited by 160
Exploring Fusion Methods for Multimodal Emotion Recognition with Missing Data
Wagner, J., Andre, E., Lingenfelser, F., & Kim, J. (2011) · IEEE Transactions on Affective Computing · doi:10.1109/t-affc.2011.12 · Cited by 156 · cited 1×
“Wanger et al. used boosting techniques to automatically determine adaptive weights for audio and visual features [10].”
MethodChristensen, J. C., Estepp, J. R., Wilson, G. F., & Russell, C. A. (2012) · NeuroImage · doi:10.1016/j.neuroimage.2011.07.091 · Cited by 144
Li, Y., Pan, J., Long, J., Yu, T., Wang, F., Yu, Z., & Wu, W. (2016) · Proceedings of the IEEE · doi:10.1109/jproc.2015.2469106 · Cited by 117
Self-citationA comparison study of two P300 speller paradigms for brain–computer interface
Pan, J., Li, Y., Gu, Z., & Yu, Z. (2013) · Cognitive Neurodynamics · doi:10.1007/s11571-013-9253-1 · Cited by 46
Self-citationPolli, E., Bersani, F. S., Rose, C. D., Liberati, D., Valeriani, G., Weisz, F., Colletti, C., Anastasia, A., & Bersani, G. (2012) · Rivista di psichiatria · doi:10.1708/1069.11717 · Cited by 46
A multi-signature brain–computer interface: use of transient and steady-state responses
Severens, M., Farquhar, J., Duysens, J., & Desain, P. (2013) · Journal of Neural Engineering · doi:10.1088/1741-2560/10/2/026005 · Cited by 36
Classification of EEG signals using feature creation produced by grammatical evolution
Tzallas, A. T., Tsoulos, I. G., Tsipouras, M. G., Γιαννακέας, Ν., Androulidakis, I. I., & Zaitseva, E. (2016) · Telecommunications Forum (TELFOR) · doi:10.1109/telfor.2016.7818809 · Cited by 10
Research works printed but not linked: 5
- 1978, Rivista di Psichiatria, 47, 126
- 1991, IEICE Transactions, 74, 3474
- 2012, WSEAS Transactions on Signal Processing, 3, 87
- 2015, Journal of Universal Computer Science, 21, 856
- 2016, IEEE Transactions on Neural Networks and Learning Systems, 28, 1241
Cited by
Qiu, S., Zhao, H., Jiang, N., Wang, Z., Liu, L., An, Y., Zhao, H., Miao, X., Liu, R., & Fortino, G. (2022) · Information Fusion · doi:10.1016/j.inffus.2021.11.006 · Cited by 671
Investigating EEG-based functional connectivity patterns for multimodal emotion recognition
Wu, X., Zheng, W.-L., Li, Z., & Lu, B.-L. (2022) · Journal of Neural Engineering · doi:10.1088/1741-2552/ac49a7 · Cited by 235
“[56] studied the fusion of EEG and facial expression data using two decision-level fusion strategies, the sum and production rules, in detecting the four basic emotions (fear, sad, happy, and neutral).”
Jiang, Y., Li, W., Hossain, M. S., Chen, M., Alelaiwi, A., & Al-Hammadi, M. (2020) · Information Fusion · doi:10.1016/j.inffus.2019.06.019 · Cited by 219
Cross-Subject Multimodal Emotion Recognition Based on Hybrid Fusion
Cimtay, Y., Ekmekcioglu, E., & Caglar-Ozhan, S. (2020) · IEEE Access · doi:10.1109/access.2020.3023871 · Cited by 200
“In [52], a decision level fusion was applied to outcomes from EEG and face modalities.”
BackgroundMethodA Review on Methods and Applications in Multimodal Deep Learning
Jabeen, S., Li, X., Amin, M. S., Bourahla, O., Li, S., & Jabbar, A. (2023) · ACM Transactions on Multimedia Computing, Communications, and Applications · doi:10.1145/3545572 · Cited by 196
“0 5 10 15 20 25 30 35 40 45 RecNet [58] MARN [59] GRU-MP/EVE [60] SibNet [61] SymSynAN [61] PickNet [69] E2E [70] RecNet(RL) [72] DPRN [73] Wei [74] M”
MethodInfluential
Metadata
- Authors
- Huang, Y., Yang, J., Liao, P., & Pan, J.
- Affiliations
- South China Normal University (CN)
- Corresponding author
- Pan, Jiahui
- Venue
- Computational Intelligence and Neuroscience
- Publisher
- Wiley
- Published
- 2017
- Volume
- 2017
- Pages
- 1-8
- ISSN
- 1687-5265 (ISSN-L)
- 1687-5273
- PMC
- PMC5625811
- OpenAlex
- W2755014019
- Semantic Scholar
- 3bab44a7b62b417ff4529325af0fba83eedf9f62
- Cited by
- 115
- References
- 20
- Citations by year
- 2026: 6 · 2025: 14 · 2024: 12 · 2023: 18 · 2022: 19 · 2021: 22 · 2020: 12 · 2019: 8 · 2018: 4
- FWCI
- 8.49
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- Computational Intelligence and Neuroscience (publishedVersion, cc-by)
- Europe PMC (PubMed Central) (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Funders
- National Natural Science Foundation of China (61503143, 2014A030310244, 201710010038)
- Natural Science Foundation of Guangdong Province (61503143, 2014A030310244, 201710010038)
- Pearl River S&T Nova Program of Guangzhou (61503143, 2014A030310244, 201710010038)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Life Sciences › Neuroscience › Cognitive Neuroscience › EEG and Brain-Computer Interfaces
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face and Expression Recognition
- Keywords
- multimodal emotion recognition, EEG, facial expressions, decision fusion, basic emotions, valence-arousal space, emotional intensity, support vector machine, neural network classifier
- MeSH terms
- Adult, Brain, Electroencephalography, Emotions, Facial Expression, Humans, Young Adult, Support Vector Machine
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.