An updated list of publications can be found in my Google Scholar profile.
- Joint Modelling of Audio-visual Cues Using Attention Mechanism for Emotion RecognitionEsam Ghaleb, Jan Niehues, and Stylianos AsteriadisMultimedia Tools and Applications 2023
Emotions play a crucial role in human-human communications with complex socio-psychological nature. In order to enhance emotion communication in human-computer interaction, this paper studies emotion recognition from audio and visual signals in video clips, utilizing facial expressions and vocal utterances. Thereby, the study aims to exploit temporal information of audio-visual cues and detect their informative time segments. Attention mechanisms are used to exploit the importance of each modality over time. We propose a novel framework that consists of bi-modal time windows spanning short video clips labeled with discrete emotions. The framework employs two networks, with each one being dedicated to one modality. As input to a modality-specific network, we consider a time-dependent signal deriving from the embeddings of the video and audio modalities. We employ the encoder part of the Transformer on the visual embeddings and another one on the audio embeddings. The research in this paper introduces detailed studies and meta-analysis findings, linking the outputs of our proposition to research from psychology. Specifically, it presents a framework to understand underlying principles of emotion recognition as functions of three separate setups in terms of modalities: audio only, video only, and the fusion of audio and video. Experimental results on two datasets show that the proposed framework achieves improved accuracy in emotion recognition, compared to state-of-the-art techniques and baseline methods not using attention mechanisms. The proposed method improves the results over baseline methods by at least 5.4%. Our experiments show that attention mechanisms reduce the gap between the entropies of unimodal predictions, which increases the bimodal predictions’ certainty and, therefore, improves the bimodal recognition rates. Furthermore, evaluations with noisy data in different scenarios are presented during the training and testing processes to check the framework’s consistency and the attention mechanism’s behavior. The results demonstrate that attention mechanisms increase the framework’s robustness when exposed to similar conditions during the training and the testing phases. Finally, we present comprehensive evaluations of emotion recognition as a function of time. The study shows that the middle time segments of a video clip are essential in the case of using audio modality. However, in the case of video modality, the importance of time windows is distributed equally.
- Dynamic Temperature Scaling in Contrastive Self-supervised Learning for Sensor-based Human Activity RecognitionBulat Khaertdinov, Stylianos Asteriadis, and Esam GhalebIEEE Transactions on Biometrics, Behavior, and Identity Science 2022
The use of deep neural networks in sensor-based Human Activity Recognition has led to considerably improved recognition rates in comparison to more traditional techniques. Nonetheless, these improvements usually rely on collecting and annotating massive amounts of sensor data, a time-consuming and expensive task. In this paper, inspired by the impressive performance of Contrastive Learning approaches in Self-Supervised Learning settings, we introduce a novel method based on the SimCLR framework and a Transformer-like model. The proposed algorithm addresses the problem of negative pairs in SimCLR by using dynamic temperature scaling within a contrastive loss function. While the original SimCLR framework scales similarities between features of the augmented views by a constant temperature parameter, our method dynamically computes temperature values for scaling. Dynamic temperature is based on instance-level similarity values extracted by an additional model pre-trained on initial instances beforehand. The proposed approach demonstrates state-of-the-art performance on three widely used datasets in sensor-based HAR, namely MobiAct, UCI-HAR and USC-HAD. Moreover, it is more robust than the identical supervised models and models trained with constant temperature in semi-supervised and transfer learning scenarios.
- Skeleton-Based Explainable Bodily Expressed Emotion Recognition Through Graph Convolutional NetworksEsam Ghaleb, André Mertens, Stylianos Asteriadis, and 1 more authorIn 2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021) 2021
Much of the focus on emotion recognition has gone into the face and voice as expressive channels, whereas bodily expressions of emotions are understudied. Moreover, current studies lack the explainability of computational features of body movements related to emotional expressions. Perceptual research on body parts’ movements shows that features related to the arms’ movements are correlated the most with human perception of emotions. In this paper, our research aims at presenting an explainable approach for bodily expressed emotion recognition. It utilizes the body joints of the human skeleton, representing them as a graph, which is used in Graph Convolutional Networks (GCNs). We improve the modelling of the GCNs by using spatial attention mechanisms based on body parts, i.e. arms, legs and torso. Our study presents a state-of-the-art explainable approach supported by experimental results on two challenging datasets. Evaluations show that the proposed methodology offers accurate performance and explainable decisions. The methodology demonstrates which body part contributes the most in its inference, showing the significance of arm movements in emotion recognition.
- Bimodal emotion recognition through audio-visual cuesEsam Ghaleb2021
Emotions play a crucial role in human-human communication with a complex socio-psychological nature, making emotion recognition a challenging task. In this dissertation, we study emotion recognition from audio and visual cues in video clips, utilizing facial expressions and speech signals, which are among the most prominent emotional expression channels. We propose novel computational methods to capture the complementary information provided by audio-visual cues for enhanced emotion recognition. The research in this dissertation shows how emotion recognition depends on emotion annotation, the perceived modalities, modalities’ robust data representations, and computational modeling. It presents progressive fusion techniques for audio-visual representations that are essential to improve their performance. Furthermore, the methods aim at exploiting the temporal dynamics of audio-visual cues and detect the informative time segments from both modalities. The dissertation presents meta-analysis studies and extensive evaluations for multimodal and temporal emotion recognition.
- Explainable and Interpretable Features of Emotion in Human Body ExpressionsAndré Mertens, Esam Ghaleb, and Stylianos AsteriadisIn BNAIC/BeneLearn 2021 2021
The cooperation between machines and humans could be improved if machines could understand and respond to the emotions of the people around them. Furthermore, the features that machines use to classify emotions should be explainable to reduce the inhibition threshold for automatic emotion recognition. However, the explainability in bodily expressivity of emotions has hardly been explored yet. Therefore, this study aims to visualize and explain the features used by neural networks to classify emotions based on body movements and postures of human characters in videos. For this purpose, a state-of-the-art neural network was selected as classification model. This network was used to classify the videos of two datasets for emotion classification. As a result, the activation of the classification features used by the model were visualized with heatmaps over the course of the videos. Furthermore, a combination of Class Activation Maps and body joint coordinates were used to compute the activation of body parts in order to investigate the existence of prototypical activation patterns in emotions. As a result, similarities were found between the activation patterns of the two datasets. These patterns may provide new insights into the classification features used by neural networks and the emotion expression in body movements and postures.
- Contrastive Self-supervised Learning for Sensor-based Human Activity RecognitionBulat Khaertdinov, Esam Ghaleb, and Stylianos AsteriadisIn 2021 IEEE International Joint Conference on Biometrics (IJCB) 2021
Deep Learning models, applied to a sensor-based Human Activity Recognition task, usually require vast amounts of annotated time-series data to extract robust features. However, annotating signals coming from wearable sensors can be a tedious and, often, not so intuitive process, that requires specialized tools and predefined scenarios, making it an expensive and time-consuming task. This paper combines one of the most recent advances in Self-Supervised Leaning (SSL), namely a SimCLR framework, with a powerful transformer-based encoder to introduce a Contrastive Self-supervised learning approach to Sensor-based Human Activity Recognition (CSSHAR) that learns feature representations from unlabeled sensory data. Extensive experiments conducted on three widely used public datasets have shown that the proposed method outperforms recent SSL models. Moreover, CSSHAR is capable of extracting more robust features than the identical supervised transformer when transferring knowledge from one dataset to another as well as when very limited amounts of annotated data are available.
- Deep Triplet Networks with Attention for Sensor-based Human Activity RecognitionBulat Khaertdinov, Esam Ghaleb, and othersIn 2021 IEEE International Conference on Pervasive Computing and Communications (PerCom) Mar 2021
One of the most significant challenges in Human Activity Recognition using wearable devices is inter-class similarities and subject heterogeneity. These problems lead to the difficulties in constructing robust feature representations that might negatively affect the quality of recognition. This study, for the first time, applies deep triplet networks with various triplet loss functions and mining methods to the Human Activity Recognition task. Moreover, we introduce a novel method for constructing hard triplets by exploiting similarities between subjects performing the same activities using the concept of Hierarchical Triplet Loss. Our deep triplet models are based on the recent state-of-the-art LSTM networks with two attention mechanisms. The extensive experiments conducted in this paper identify important hyperparameters and settings for training deep metric learning models on widely-used open-source Human Activity Recognition datasets. The comparison of the proposed models against the recent benchmark models shows that deep metric learning approach has the potential to improve the quality of recognition. Specifically, at least one of the implemented triplet networks shows the state-of-the-art results for each dataset used in this study, namely PAMAP2, USC-HAD and MHEALTH. Another positive effect of applying deep triplet networks and especially the proposed sampling algorithm is that feature representations are less affected by inter-class similarities and subject heterogeneity issues.
- BNAIC20Deep, dimensional and multimodal emotion recognition using attention mechanismsJan Lucas, Esam Ghaleb, and Stylianos AsteriadisIn BNAIC/BeneLearn 2020 Mar 2020
Emotion recognition is an increasingly important sub-field in artificial intelligence (AI). Advances in this field could drastically change the way people interact with computers and allow for automation of tasks that currently require a lot of manual work. For example, registering the emotion a subject expresses for a potential advert. Previous work has shown that using multiple modalities, although challenging, is very beneficial. Affective cues in audio and video may not occur simultaneously, and the modalities do not always contribute equally to emotion. This work seeks to apply attention mechanisms to aid in the fusion of audio and video, for the purpose of emotion recognition using state-of-the-art techniques from artificial intelligence and, more specifically, deep neural networks. To achieve this, two forms of attention are used. Embedding attention applies attention on the input of a modalityspecific model, allowing recurrent networks to consider multiple input time steps. Bimodal attention fusion applies attention to fuse the output of modality-specific networks. Combining both these attention mechanisms yielded CCCs of 0.62 and 0.72 for arousal and valence respectively on the RECOLA dataset used in AVEC 2016. These results are competitive with the state-of-the-art, underlying the potential of attention mechanisms in multimodal fusion for behavioral signals.
- Multimodal Attention-Mechanism For Temporal Emotion RecognitionEsam Ghaleb, Jan Niehues, and Stylianos AsteriadisIn 2020 IEEE International Conference on Image Processing (ICIP) Mar 2020
Exploiting the multimodal and temporal interaction between audio-visual channels is essential for automatic audio-video emotion recognition (AVER). Modalities’ strength in emotions and time-window of a video-clip could be further utilized through a weighting scheme such as attention mechanism to capture their complementary information. The attention mechanism is a powerful approach for sequence modeling, which can be employed to fuse audio-video cues overtime. We propose a novel framework which consists of biaudio-visual time-windows that span short video-clips labeled with discrete emotions. Attention is used to weigh these time windows for multimodal learning and fusion. Experimental results on two datasets show that the proposed methodology can achieve an enhanced multimodal emotion recognition.
- Temporal triplet mining for personality recognitionDario Dotti, Esam Ghaleb, and Stylianos AsteriadisIn 2020 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2020) Mar 2020
One of the primary goals of personality computing is to enhance the automatic understanding of human behavior, making use of various sensing technologies. Recent studies have started to correlate personality patterns described by psychologists with data findings, however, given the subtle delineations of human behaviors, results are specific to predefined contexts. In this paper, we propose a framework for automatic personality recognition that is able to embed different behavioral dynamics evoked by diverse real world scenarios. Specifically, motion features are designed to encode local motion dynamics from the human body, and interpersonal distance (proxemics) features are designed to encode global dynamics in the scene. By using a Convolutional Neural Network (CNN) architecture which utilizes a triplet loss deep metric learning, we learn temporal, as well as discriminative spatio-temporal streams of embeddings to represent patterns of personality behaviors. We experimentally show that the proposed Temporal Triplet Mining strategy leverages the similarity between temporally related samples and, therefore, helps to encode higher semantic movements or sub-movements which are easier to map onto personality labels. Our experiments show that the generated embeddings improve the state-of-the-art results of personality recognition on two public datasets, recorded in different scenarios.
- Multimodal and temporal perception of audio-visual cues for emotion recognitionEsam Ghaleb, Mirela Popa, and Stylianos AsteriadisIn 2019 8th International Conference on Affective Computing and Intelligent Interaction (ACII) Mar 2019
In Audio-Video Emotion Recognition (AVER), the idea is to have a human-level understanding of emotions from video clips. There is a need to bring these two modalities into a unified framework, to effectively learn multimodal fusion for AVER. In addition, literature studies lack in-depth analysis and utilization of how emotions vary as a function of time. Psychological and neurological studies show that negative and positive emotions are not recognized at the same speed. In this paper, we propose a novel multimodal temporal deep network framework that embeds video clips using their audio-visual content, onto a metric space, where their gap is reduced and their complementary and supplementary information is explored. We address two research questions, (1) how audio-visual cues contribute to emotion recognition and (2) how temporal information impacts the recognition rate and speed of emotions. The proposed method is evaluated on two datasets, CREMA-D and RAVDESS. The study findings are promising, achieving the state-of-the-art performance on both datasets, and showing a significant impact of multimodal and temporal emotion perception.
- Metric learning-based multimodal audio-visual emotion recognitionEsam Ghaleb, Mirela Popa, and Stylianos AsteriadisIeee Multimedia Mar 2019
People express their emotions through multiple channels, such as visual and audio ones. Consequently, automatic emotion recognition can be significantly benefited by multimodal learning. Even-though each modality exhibits unique characteristics; multimodal learning takes advantage of the complementary information of diverse modalities when measuring the same instance, resulting in enhanced understanding of emotions. Yet, their dependencies and relations are not fully exploited in audio–video emotion recognition. Furthermore, learning an effective metric through multimodality is a crucial goal for many applications in machine learning. Therefore, in this article, we propose multimodal emotion recognition metric learning (MERML), learned jointly to obtain a discriminative score and a robust representation in a latent-space for both modalities. The learned metric is efficiently used through the radial basis function (RBF) based support vector machine (SVM) kernel. The evaluation of our framework shows a significant performance, improving the state-of-the-art results on the eNTERFACE and CREMA-D datasets.
- Exploiting sensing devices availability in AR/VR deployments to foster engagementNicholas Vretos, Petros Daras, Stylianos Asteriadis, and 7 more authorsVirtual Reality Mar 2019
Currently, in all augmented reality (AR) or virtual reality (VR) educational experiences, the evolution of the experience (game, exercise or other) and the assessment of the user’s performance are based on her/his (re)actions which are continuously traced/sensed. In this paper, we propose the exploitation of the sensors available in the AR/VR systems to enhance the current AR/VR experiences, taking into account the users’ affect state that changes in real time. Adapting the difficulty level of the experience to the users’ affect state fosters their engagement which is a crucial issue in educational environments and prevents boredom and anxiety. The users’ cues are processed enabling dynamic user profiling. The detection of the affect state based on different sensing inputs, since diverse sensing devices exist in different AR/VR systems, is investigated, and techniques that have been undergone validation using state-of-the-art sensors are presented.
- Towards Affect Recognition through Interactions with Learning MaterialsEsam Ghaleb, Mirela Popa, Enrique Hortal, and 2 more authorsIn Machine Learning and Applications (ICMLA), 2018 17th International Conference on Mar 2018
Affective state recognition has recently attracted a notable amount of attention in the research community, as it can be directly linked to a student’s performance during learning. Consequently, being able to retrieve the affect of a student can lead to more personalized education, targeting higher degrees of engagement and, thus, optimizing the learning experience and its outcomes. In this paper, we apply Machine Learning (ML) and present a novel approach for affect recognition in Technology-Enhanced Learning (TEL) by understanding learners’ experience through tracking their interactions with a serious game as a learning platform. We utilize a variety of interaction parameters to examine their potential to be used as an indicator of the learner’s affective state. Driven by the Theory of Flow model, we investigate the correspondence between the prediction of users’ self-reported affective states and the interaction features. Cross-subject evaluation using Support Vector Machines (SVMs) on a dataset of 32 participants interacting with the platform demonstrated that the proposed framework could achieve a significant precision in affect recognition. The subject-based evaluation highlighted the benefits of an adaptive personalized learning experience, contributing to achieving optimized levels of engagement.
- DEEP REPRESENTATION AND SCORE NORMALIZATION FOR FACE RECOGNITION UNDER MISMATCHED CONDITIONSEsam Ghaleb, Gokhan Ozbulak, Hua Gao, and 1 more authorMar 2018
Face recognition under unconstrained conditions is a challenging computer vision task. Identification under mismatched conditions, for example, due to difference of view angles, illumination conditions, and image quality between galley and probe images, as in the International Challenge on Biometric Recognition-in-the-Wild (ICB-RW) 2016, poses even further challenges. In our work, to address this problem, we have employed facial image preprocessing, deep representation, and score normalization methods to develop a successful face recognition system. In the preprocessing step, we have aligned the gallery and probe face images with respect to automatically detected eye centers. We only used frontal faces as a gallery. For face representation, we have employed a state-of-the-art deep convolutional neural network model, namely the VGGFace model. For classification, we have applied a nearest neighbor classifier with correlation distance as the distance metric. As the final step, we normalized the resulting similarity score matrix, which includes the scores of all face images in the probe set against all face images in the gallery set, with z-score normalization. The proposed system has achieved 69.8 percent Rank-1 and 85.3 percent Rank-5 accuracy on the test set, which were the highest accuracies obtained in the challenge.
- High-performance and lightweight real-time deep face emotion recognitionJustus Schwan, Esam Ghaleb, Enrique Hortal, and 1 more authorIn Semantic and Social Media Adaptation and Personalization (SMAP), 2017 12th International Workshop on Mar 2017
Deep learning is used for all kinds of tasks which require human-like performance, such as voice and image recognition in smartphones, smart home technology, and self-driving cars. While great advances have been made in the field, results are often not satisfactory when compared to human performance. In the field of facial emotion recognition, especially in the wild, Convolutional Neural Networks (CNN) are employed because of their excellent generalization properties. However, while CNNs can learn a representation for certain object classes, an amount of (annotated) training data roughly proportional to the class’s complexity is needed and seldom available. This work describes an advanced pre-processing algorithm for facial images and a transfer learning mechanism, two potential candidates for relaxing this requirement. Using these algorithms, a lightweight face emotion recognition application for Human-Computer Interaction with TurtleBot units was developed.
- Multimodal fusion based on information gain for emotion recognition in the wildEsam Ghaleb, Mirela Popa, Enrique Hortal, and 1 more authorIn Intelligent Systems Conference (IntelliSys), 2017 Mar 2017
- Accio: A data set for face track retrieval in movies across ageEsam Ghaleb, Makarand Tapaswi, Ziad Al-Halah, and 2 more authorsIn Proceedings of the 5th ACM on International Conference on Multimedia Retrieval Mar 2015
In this paper we present a novel approach towards multi-modal emotion recognition on a challenging dataset AFEW’16, composed of video clips labeled with the six basic emotions plus the neutral state. After a preprocessing stage, we employ different feature extraction techniques (CNN, DSIFT on face and facial ROI, geometric and audio based) and encoded frame-based features using Fisher vector representations. Next, we leverage the properties of each modality using different fusion schemes. Apart from the early-level fusion and the decision level fusion approaches, we propose a hierarchical decision level method based on information gain principles and we optimize its parameters using genetic algorithms. The experimental results prove the suitability of our method, as we obtain 53.06% validation accuracy, surpassing by 14% the baseline of 38.81% on a challenging dataset, suitable for emotion recognition in the wild.
- A Face Recognition Based Multiplayer Mobile Game ApplicationUgur Demir, Esam Ghaleb, and Hazım Kemal EkenelIn IFIP International Conference on Artificial Intelligence Applications and Innovations Mar 2014
In this paper, we present a multiplayer mobile game application that aims at enabling individuals play paintball or laser tag style games using their smartphones. In the application, face detection and recognition technologies are utilised to detect and identify the individuals, respectively. In the game, first, one of the players starts the game and invites the others to join. Once everyone joins the game, they receive a notification for the training stage, at which they need to record another player’s face for a short time. After the completion of the training stage, the players can start shooting each other, that is, direct the smartphone to another user and when the face is visible, press the shoot button on the screen. Both the shooter and the one who is shot are notified by the system after a successful hit. To realise this game in real-time, fast and robust face detection and face recognition algorithms have been employed. The face recognition performance of the system is benchmarked on the face data collected from the game, when it is played with up to ten players. It is found that the system is able to identify the players with a success rate of around or over 90% depending on the number of players in the game.
- An energy efficient routing technique and implementation in WSNsAbdullah Aydeger, Esam Ghaleb, and Sema OktugIn Signal Processing and Communications Applications Conference (SIU), 2014 22nd Mar 2014
In this work, an energy efficient routing technique for WSNs is introduced. Here the routes between the source nodes and the base station are formed considering the energy levels of the intermediate nodes. These routes are refreshed dynamically to recover topology changes and to keep the energy level of the nodes close to each other. The technique introduced together with data aggregation are implemented on the WSN testbed at the department of computer enngineering, ITU. The results obtained have shown that the routing technique introduced and data aggregation contribute to the lifetime of WSNs.