Multimodal emotion recognition using deep learning architectures
The DOI matches its Crossref record; 29 of the 39 references it cites were found in a registry or an index; none of the 12 sources its claims cite is checked for support yet.
Abstract
Emotion analysis and recognition has become an interesting topic of research among the computer vision research community. In this paper, we first present the emoF-BVP database of multimodal (face, body gesture, voice and physiological signals) recordings of actors enacting various expressions of emotions. The database consists of audio and video sequences of actors displaying three different intensities of expressions of 23 different emotions along with facial feature tracking, skeletal tracking and the corresponding physiological data. Next, we describe four deep belief network (DBN) models and show that these models generate robust multimodal features for emotion classification in an unsupervised manner. Our experimental results show that the DBN models perform better than the state of the art methods for emotion recognition. Finally, we propose convolutional deep belief network (CDBN) models that learn salient multimodal features of expressions of emotions. Our CDBN models give better recognition accuracies when recognizing low intensity or subtle expressions of emotions when compared to state of the art methods.
Overview
CoChat Summary
TL;DR
The paper introduces the emoFBVP multimodal emotion database (face, body, voice, physiological signals) and evaluates deep belief network (DBN) and convolutional DBN (CDBN) models for emotion recognition. It reports that DBN models outperform SVM baselines on emoFBVP and standard corpora, and that CDBN models with region-of-interest extraction improve recognition of low-intensity/subtle expressions.
Key findings
- The emoFBVP database contains recordings from 10 professional actors displaying 23 emotions in three intensities, with six examples per emotion and 1,380 samples of audio, face/body video, and physiological data, plus facial and skeletal tracking.
- For DemoFBVP models, 3DemoFBVP achieved 90.10% ideal-data accuracy vs. baseline IG 89.41% and PFA 85.33%; non-ideal accuracy was 73.11% vs. 68.89% and 68.71%; combined-data accuracy was 83.10% for DemoFBVP+f vs. 79.82% and 79.90%.
- For low-intensity/subtle expressions, CDemoFBVP+ROI reached 83.18% on emoFBVP vs. CDemoFBVP 81.41%, DemoFBVP 76.54%, and SVM baseline 75.67%; on MAHNOB-HCI it reached 58.5% vs. 57.9%, 53.1%, and 52.4%.
- On standard corpora, deep models outperformed SVM baselines: Cohn Kanade 3DemoF 96.3% vs. 95.4%; Mind Reading 3DemoV 92.87% vs. 90.62%; DEAP 3DemoP 79.2% vs. 78.6%; MAHNOB-HCI 3DemoFBVP 54.8% vs. 52.4%.
Methods
- Data capture: 10 professional actors expressed 23 emotions in three intensities across standing and seated sessions, with six examples per emotion; recordings used Microsoft Kinect, Brekel Kinect Pro Face/Body, Zephyr BioHarness, and wrist-worn accelerometers, yielding 1,380 samples labeled by three evaluators and split into ideal, non-ideal, and combined data.
- Features: prosodic and spectral voice features (pitch, energy, mel-frequency filter banks), facial tracking, skeletal tracking, and physiological signals were summarized with mean, variance, lower/upper quantiles, and quantile range, giving 180 voice + 540 face + 540 body + 120 physiological = 1,380 features, normalized to avoid person dependency.
- Models/analysis: four DBN DemoDBN variants (DemoFBVP, f+DemoFBVP, DemoFBVP+f, 3DemoFBVP) plus bimodal variants; CDBN models built from convolutional RBMs with probabilistic max-pooling; SVM baselines with IG or PFA feature selection, RBF kernels, one-versus-all training, and leave-one-person-out cross validation, evaluated on emoFBVP and standard emotion corpora.
Limitations
- The paper does not explicitly state limitations; evident from the text, the expressions are acted by 10 professional actors rather than spontaneous, and the database contains 1,380 samples.
- Evident from the results: low-intensity/subtle expression recognition is lower than ideal-data recognition (e.g., 3DemoFBVP non-ideal 73.11% vs. ideal 90.10%).
- Evident from the text: real-time multimodal recognition and robustness when one or more modalities are absent are stated as future work, not demonstrated.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 2 Not valid · 9 Not checked · 27 Own results and interpretations
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
“we describe four deep belief network (DBN) models and show that these models generate robust multimodal features for emotion classifica- tion in an unsupervised manner.”
- Also stated in: § 1. Introduction · plain
“We describe four deep belief network (DBN) mod- els and show that they generate robust multimodal features for emotion classification in an unsupervised manner.”
plainstates 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.
- Also stated in: § 1. Introduction · 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
“This database is comprised of 1380 samples of audio se- quences, video sequences of face and body and physiolog- ical data corresponding to various expressions of emotions.”
§ 3. emoFBVP Database > 3.1. Properties of emoFBVP Database
plainn = 1,380- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, the database provides six examples of each of the 23 emotions in varying intensities of expression.”
plainn = 10- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“we propose convolutional deep belief network (CDBN) mod- els that learn salient multimodal features of expressions of emotions.”
plainstates 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.
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
“Our CDBN models give better recognition accu- racies when recognizing low intensity or subtle expressions of emotions when compared to state of the art methods.”
plainstates 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.
13 more claims
Create a free account to see all 38 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 11
- It may state two separate claims as one. 5
- The quote only partly states this claim. 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 cited source says what the claim attributes to it.
In the citing paper
“Kim et al. [159] and others (e.g., [149], [148]) demonstrated how this can be achieved with DBNs.”
From Deep learning for human affect recognition: Insights and new developments (2021) · cites this paper as [149]
In the citing paper
“For example, Zhang et al. [148] and Ranganathan et al. [149] used fully-connected DBN models to learn from aggregated facial feature point trajectories, both improving recognition accuracies over shallow aggregation strategies.”
From Deep learning for human affect recognition: Insights and new developments (2021) · cites this paper as [149]
In the citing paper
“Ranganathan et al. [149] demonstrated the feasibility of learning joint feature representations of AVP sensor data with approach J1 and a DBN, but do not compare the performances of different modality combinations.”
From Deep learning for human affect recognition: Insights and new developments (2021) · cites this paper as [149]
Citations
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
A fast learning algorithm for deep belief nets
Hinton, G. E., Osindero, S., & Teh, Y.-W. (2006) · Neural Computation · doi:10.1162/neco.2006.18.7.1527 · Cited by 16,336
Learning Deep Architectures for AI
Bengio, Y. (2009) · Foundations and Trends® in Machine Learning · doi:10.1561/2200000006 · Cited by 9,239
Learning Deep Architectures for AI
Bengio, Y. (2009) · now publishers, Inc. eBooks · doi:10.1561/9781601982957 · Cited by 4,941
Training Products of Experts by Minimizing Contrastive Divergence
Hinton, G. E. (2002) · Neural Computation · doi:10.1162/089976602760128018 · Cited by 4,879 · cited 2×
“For training the CRBMs, we use contrastive divergence [23] to approximate the gradient of the log-likelyhood term effectively.”
Method(2011) · Cited by 2,752
Lee, H., Grosse, R., Ranganath, R., & Ng, A. Y. (2009) · Proceedings of the 26th Annual International Conference on Machine Learning · doi:10.1145/1553374.1553453 · Cited by 2,412
Information processing in dynamical systems: foundations of harmony theory
Smolensky, P. (1986) · The MIT Press eBooks · Cited by 2,232
Multimodal Learning with Deep Boltzmann Machines
Srivastava, N., & Salakhutdinov, R. (2012) · Journal of Machine Learning Research · doi:10.5555/2627435.2697059 · Cited by 1,807
Acoustic Modeling Using Deep Belief Networks
Mohamed, A.-R., Dahl, G. E., & Hinton, G. (2012) · IEEE Transactions on Audio, Speech, and Language Processing · doi:10.1109/tasl.2011.2109382 · Cited by 1,727
Research works printed but not linked: 1
- desjardins, 2008, Empirical evaluation of convolutional RBMs for vision, Universite De Montreal Tech Rep
Cited by
Items 201–238 of 238
Multi-Layer Progressive Face Alignment by Integrating Global Match and Local Refinement
Gao, N., Wang, X., & Wang, X. (2019) · Applied Sciences · doi:10.3390/app9050977 · Cited by 1
Multi-Modal Based Fusion Emotion Identification Using Inception V3 Algorithm
P, S. V., Srivarshini., R., & Grace, R. K. (2023) · International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS) · doi:10.1109/icssas57918.2023.10331693 · Cited by 1
VATE: A Large Scale Multimodal Spontaneous Dataset for Affective Evaluation
Agnelli, F., Grossi, G., D’Amelio, A., Paoli, M. D., & Lanzarotti, R. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-91575-8_13 · Cited by 2
Review on recent trends and challenges in deep learning and artificial intelligence
Shuaib, M., Jaafar, J., Faizuddin, M., & Musa, S. (2022) · AIP conference proceedings · doi:10.1063/5.0119678 · Cited by 2
Chernyshеv, D., Mihaylenko, V., & Tereikovska, L. (2020) · Management of Development of Complex Systems · doi:10.32347/2412-9933.2020.43.123-130 · Cited by 1
Vala, J., & Jaliya, U. K. (2023) · International Journal of Engineering Applied Sciences and Technology · doi:10.33564/ijeast.2023.v08i04.021 · Cited by 1
Singh, A. K., & Shukla, A. K. (2025) · IEEE International Conference on Intelligent Signal Processing and Effective Communication Technologies (INSPECT) · doi:10.1109/inspect67393.2025.11350529 · Cited by 1
Human Emotion Detection Using Deep Learning Neural Network
Rao, N., Tanveer, S., Khalid, S. S., & Alam, N. (2022) · Lecture notes in networks and systems · doi:10.1007/978-981-19-2535-1_42 · Cited by 1
Singh, A., & Shukla, A. K. (2026) · Journal of The Institution of Engineers (India) Series B · doi:10.1007/s40031-026-01340-6
Steering Data Arbitration on Facial-Speech Features for Fusion-Based Emotion Recognition Framework
Singh, V., & Singh, K. (2023) · Lecture notes in networks and systems · doi:10.1007/978-981-99-3243-6_9
Efficient Selective Labeling for Few-Shot 3D Point Cloud Semantic Segmentation
Zheng, X., Chen, N., Pan, Z., Zou, R., & Song, X. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-22261-9_32
Letaifa, L. B., Bohi, A., & Slama, R. (2025) · Journal on Multimodal User Interfaces · doi:10.1007/s12193-025-00468-x · Cited by 1
Improving information fusion in deep learning
Don, D. P. M. K. (2023) · Queensland University of Technology · doi:10.5204/thesis.eprints.241291
Beyond Words: Enhancing Desire, Emotion, and Sentiment Recognition with Non-Verbal Cues
Chen, W., Wang, T., Xue, F., Li, J., Liu, H., & Sha, Y. (2026) · International World Wide Web Conference (WWW) · doi:10.1145/3774904.3792393 · Cited by 1
OratorPath: An AI-Powered Framework for Enhanced Public Speaking Proficiency
Aizaz, F., Ehsan, L., Tariq, M. T., & Rehman, S. U. (2026) · International Journal of Innovations in Science and Technology · doi:10.33411/ijist/1841
Video-Based Contactless Mood Detection Combining Heart Rate and Facial Expressions
Manni, A., Caroppo, A., Leone, A., & Siciliano, P. A. (2022) · Lecture notes in electrical engineering · doi:10.1007/978-3-031-08136-1_48
Multimodal Emotion Detection and Classification of Audio-Visual Using Deep Learning
Srivastava, M., Singh, P., & Sharma, M. K. (2025) · International Conference on Advanced Computing and Emerging Technologies (ACET) · doi:10.1109/acet67282.2025.11430197
Enhancing Sentiment Analysis Accuracy Through Multimodal Data Fusion: A Deep Learning Approach
Tiwari, S., Jain, A., & Mishra, A. (2024) · Lecture notes in networks and systems · doi:10.1007/978-981-97-6726-7_45
Investigating the Impact of Parameter Variation on Speech Emotion Analysis Using LSTM Models
Odedra, R., Bhoi, D., Bhatt, N., & Thakkar, A. R. (2025) · Lecture notes in electrical engineering · doi:10.1007/978-981-96-4391-2_8
Speech emotion recognition for audio and text
Kumar, J., Kumar, G., Rao, P. N., Philip, J., & Prashanth, M. (2023) · AIP conference proceedings · doi:10.1063/5.0113389
A Study on Multimodal Foundation Models for Affective Video Prediction
Agnelli, F., Ditroia, M., Blandano, G., D’Amelio, A., Ghezzi, O., Paoli, M. D., & Lanzarotti, R. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-10192-1_19
Deep Active Learning Explored Across Diverse Label Spaces
Ranganathan, H. (2018)
Self-citation3D Skeletal Movement enhanced Emotion Recognition Network
Shi, J., Liu, C., Ishi, C. T., & Ishiguro, H. (2020) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)
EmotiGuard: Real-Time Human Emotion Recognition System for Adult Monitoring
Apurupa, M., & Sngh, H. (2024) · International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication (IConSCEPT) · doi:10.1109/iconscept61884.2024.10627881
Non-contact Emotion Recognition using Doppler Radar
Valdezco, J. A. T., & Balbin, E. J. J. R. (2025) · IEEE International Conference on Internet of Things and Intelligence Systems (IoTaIS) · doi:10.1109/iotais67227.2025.11282132
Ilyas, C. M. A. (2021) · VBN Forskningsportal (Aalborg Universitet) · doi:10.54337/aau460118975
HCC: An Explainable Framework for Classifying Discomfort from Video
Valentine, W., Webb, M., Collum, C., Feil-Seifer, D., & Hand, E. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-77389-1_23
Student Behavior Analysis in Classroom Using Facial Emotion Recognitions and Graph Visualizations
Arsath, J. A., Narayanan, N. P., Abirami, A. M., & Karthikeyan, P. (2024) · Lecture notes in networks and systems · doi:10.1007/978-981-97-5412-0_16
Learning multimodal interaction models in mixed societies
Malik, M. U. (2020) · theses.fr (ABES) · doi:10.70675/25fc786ez7f06z47dcz93b9z7439be248efb · Cited by 1
Multimodal data analysis in emotion recognition: a review
Berdyshev, D. A., & Shishkin, A. G. (2025) · Modeling and Analysis of Information Systems · doi:10.18255/1818-1015-2025-3-252-281 · Cited by 1
Face Emotion Recognition (FER) Using Convolutional Neural Network (CNN) in Machine Learning
Jain, L., & Kavita, K. (2024) · International Journal for Research in Applied Science and Engineering Technology · doi:10.22214/ijraset.2024.58077
Kong, X., & Yu, W. (2025) · International Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS) · doi:10.1109/iciibms66230.2025.11316745
SIREN: Scalable isotropic recursive column multimodal neural architecture
Truchan, H., Farag, A., & Ahmadi, Z. (2025) · Information Fusion · doi:10.1016/j.inffus.2025.103500
Development of Techniques for Speech Emotion Recognition (SER) In the Context of Deep Learning
Triandi, B., Mawengkang, H., Efendi, S., & Syawaluddin (2022) · International Conference on Cybernetics and Intelligent System (ICORIS) · doi:10.1109/icoris56080.2022.10031385
Metadata
- Authors
- Ranganathan, H., Chakraborty, S., & Panchanathan, S.
- Affiliations
- Arizona State University (US)
- Venue
- 2016 IEEE Winter Conference on Applications of Computer Vision (WACV)
- Publisher
- IEEE
- Published
- 2016-03
- Pages
- 1-9
- ISSN
- 2472-6737 (ISSN-L)
- OpenAlex
- W2406223855
- Semantic Scholar
- cf128d911db07a162b7f32bfb94ca62ab5d737ca
- Cited by
- 249
- References
- 39
- Citations by year
- 2026: 8 · 2025: 31 · 2024: 30 · 2023: 36 · 2022: 40 · 2021: 15 · 2020: 29 · 2019: 27 · 2018: 20 · 2017: 11 · 2016: 2
- FWCI
- 23.13
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE Winter Conference on Applications of Computer Vision (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face and Expression Recognition
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face recognition and analysis
- Keywords
- multimodal emotion recognition, deep belief network, convolutional deep belief network, facial expression intensity, physiological signals, body gestures, voice, facial feature tracking, skeleton tracking, unsupervised feature learning
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.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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.