Nancy ZlatintsiPostdoctoral Research Associate
(+30) 210-772-2964
- nzlat@cs.ntua.gr
- Office 2.2.19
Biosketch
I was born in Thessaloniki, Greece and received my Diploma degree in Media Engineering from the Royal Institute of Technology (KTH), Stockholm, Sweden in September 2006. My Master’s degree thesis, which was conducted at the Department of Speech, Music and Hearing (TMH – Fant Laboratorium), under the supervision of Kjetil Falkenberg Hansen and Prof. Anders Askenfelt, was in music acoustics and specifically the sound of clarinet. This thesis was an unofficial part of the European project VEMUS.
Since January 2007 I was a Ph.D. candidate in CVSP group – at the Electrical and Computer Engineering School in the National Technical University of Athens, Greece – under the supervision of Prof. Petros Maragos in the general areas of Audio and Multimedia Processing. During this period, I partially participated in the EU MUSCLE project, with human movie annotation and human evaluations. My research interests lie in the areas of music information retrieval and audio processing and include analysis and recognition.
In December 2013 I received my PhD degree with title “Music Signal Processing and Applications in Recognition”. I currently work as a Postdoctoral Research Associate in CVSP group in related topics and I paricipate in various European and Greek research programs.
I have also studied musicology for four semesters at Stockholm University in Sweden. It is not hard to understand that when not studying my biggest obsession is music – listening, reading, playing and talking about it. Some of my other interests are photography and photo editing, books, movies and travelling.
For my doctoral studies in the area of MIR since February 2011, my research has been co-financed by the European Union (European Social Fund – ESF) and Greek national funds through the Operational Program “Education and Lifelong Learning” of the National Strategic Reference Framework (NSRF) – Research Funding Program: Heracleitus II. Investing in knowledge society through the European Social Fund.
Recent Research Projects
Publications
2018 |
G Bouritsas, P Koutras, A Zlatintsi, Petros Maragos Multimodal Visual Concept Learning with Weakly Supervised Techniques Conference Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, Utah, USA, 2018. Abstract | BibTeX | Links: [PDF] @conference{BKA+18, title = {Multimodal Visual Concept Learning with Weakly Supervised Techniques}, author = {G Bouritsas and P Koutras and A Zlatintsi and Petros Maragos}, url = {http://robotics.ntua.gr/wp-content/uploads/sites/2/2018_BKZM_MultimodalVisualConceptLearningWeaklySupervisedTechniques_CVPR.pdf}, year = {2018}, date = {2018-06-01}, booktitle = {Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, address = { Salt Lake City, Utah, USA}, abstract = {Despite the availability of a huge amount of video data accompanied by descriptive texts, it is not always easy to exploit the information contained in natural language in order to automatically recognize video concepts. Towards this goal, in this paper we use textual cues as means of supervision, introducing two weakly supervised techniques that extend the Multiple Instance Learning (MIL) framework: the Fuzzy Sets Multiple Instance Learning (FSMIL) and the Probabilistic Labels Multiple Instance Learning (PLMIL). The former encodes the spatio-temporal imprecision of the linguistic descriptions with Fuzzy Sets, while the latter models different interpretations of each description’s semantics with Probabilistic Labels, both formulated through a convex optimization algorithm. In addition, we provide a novel technique to extract weak labels in the presence of complex semantics, that consists of semantic similarity computations. We evaluate our methods on two distinct problems, namely face and action recognition, in the challenging and realistic setting of movies accompanied by their screenplays, contained in the COGNIMUSE database. We show that, on both tasks, our method considerably outperforms a state-of-the-art weakly supervised approach, as well as other baselines.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Despite the availability of a huge amount of video data accompanied by descriptive texts, it is not always easy to exploit the information contained in natural language in order to automatically recognize video concepts. Towards this goal, in this paper we use textual cues as means of supervision, introducing two weakly supervised techniques that extend the Multiple Instance Learning (MIL) framework: the Fuzzy Sets Multiple Instance Learning (FSMIL) and the Probabilistic Labels Multiple Instance Learning (PLMIL). The former encodes the spatio-temporal imprecision of the linguistic descriptions with Fuzzy Sets, while the latter models different interpretations of each description’s semantics with Probabilistic Labels, both formulated through a convex optimization algorithm. In addition, we provide a novel technique to extract weak labels in the presence of complex semantics, that consists of semantic similarity computations. We evaluate our methods on two distinct problems, namely face and action recognition, in the challenging and realistic setting of movies accompanied by their screenplays, contained in the COGNIMUSE database. We show that, on both tasks, our method considerably outperforms a state-of-the-art weakly supervised approach, as well as other baselines. |
A Zlatintsi, I Rodomagoulakis, P Koutras, A ~C Dometios, V Pitsikalis, C ~S Tzafestas, P Maragos Multimodal Signal Processing and Learning Aspects of Human-Robot Interaction for an Assistive Bathing Robot Conference Proc. IEEE Int'l Conf. Acous., Speech, and Signal Processing, Calgary, Canada, 2018. Abstract | BibTeX | Links: [PDF] @conference{ZRK+18, title = {Multimodal Signal Processing and Learning Aspects of Human-Robot Interaction for an Assistive Bathing Robot}, author = {A Zlatintsi and I Rodomagoulakis and P Koutras and A ~C Dometios and V Pitsikalis and C ~S Tzafestas and P Maragos}, url = {http://robotics.ntua.gr/wp-content/publications/Zlatintsi+_I-SUPPORT_ICASSP18.pdf}, year = {2018}, date = {2018-04-01}, booktitle = {Proc. IEEE Int'l Conf. Acous., Speech, and Signal Processing}, address = {Calgary, Canada}, abstract = {We explore new aspects of assistive living on smart human-robot interaction (HRI) that involve automatic recognition and online validation of speech and gestures in a natural interface, providing social features for HRI. We introduce a whole framework and resources of a real-life scenario for elderly subjects supported by an assistive bathing robot, addressing health and hygiene care issues. We contribute a new dataset and a suite of tools used for data acquisition and a state-of-the-art pipeline for multimodal learning within the framework of the I-Support bathing robot, with emphasis on audio and RGB-D visual streams. We consider privacy issues by evaluating the depth visual stream along with the RGB, using Kinect sensors. The audio-gestural recognition task on this new dataset yields up to 84.5%, while the online validation of the I-Support system on elderly users accomplishes up to 84% when the two modalities are fused together. The results are promising enough to support further research in the area of multimodal recognition for assistive social HRI, considering the difficulties of the specific task.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } We explore new aspects of assistive living on smart human-robot interaction (HRI) that involve automatic recognition and online validation of speech and gestures in a natural interface, providing social features for HRI. We introduce a whole framework and resources of a real-life scenario for elderly subjects supported by an assistive bathing robot, addressing health and hygiene care issues. We contribute a new dataset and a suite of tools used for data acquisition and a state-of-the-art pipeline for multimodal learning within the framework of the I-Support bathing robot, with emphasis on audio and RGB-D visual streams. We consider privacy issues by evaluating the depth visual stream along with the RGB, using Kinect sensors. The audio-gestural recognition task on this new dataset yields up to 84.5%, while the online validation of the I-Support system on elderly users accomplishes up to 84% when the two modalities are fused together. The results are promising enough to support further research in the area of multimodal recognition for assistive social HRI, considering the difficulties of the specific task. |
2017 |
A Zlatintsi, P Koutras, G Evangelopoulos, N Malandrakis, N Efthymiou, K Pastra, A Potamianos, P Maragos COGNIMUSE: a multimodal video database annotated with saliency, events, semantics and emotion with application to summarization Journal Article EURASIP Journal on Image and Video Processing, 54 , pp. 1–24, 2017. Abstract | BibTeX | Links: [PDF] @article{ZKE+17, title = {COGNIMUSE: a multimodal video database annotated with saliency, events, semantics and emotion with application to summarization}, author = {A Zlatintsi and P Koutras and G Evangelopoulos and N Malandrakis and N Efthymiou and K Pastra and A Potamianos and P Maragos}, url = {http://robotics.ntua.gr/wp-content/publications/Zlatintsi+_COGNIMUSEdb_EURASIP_JIVP-2017.pdf}, doi = {doi 10.1186/s13640-017-0194}, year = {2017}, date = {2017-01-01}, journal = {EURASIP Journal on Image and Video Processing}, volume = {54}, pages = {1--24}, abstract = {Research related to computational modeling for machine-based understanding requires ground truth data for training, content analysis, and evaluation. In this paper, we present a multimodal video database, namely COGNIMUSE, annotated with sensory and semantic saliency, events, cross-media semantics, and emotion. The purpose of this database is manifold; it can be used for training and evaluation of event detection and summarization algorithms, for classification and recognition of audio-visual and cross-media events, as well as for emotion tracking. In order to enable comparisons with other computational models, we propose state-of-the-art algorithms, specifically a unified energy-based audio-visual framework and a method for text saliency computation, for the detection of perceptually salient events from videos. Additionally, a movie summarization system for the automatic production of summaries is presented. Two kinds of evaluation were performed, an objective based on the saliency annotation of the database and an extensive qualitative human evaluation of the automatically produced summaries, where we investigated what composes high-quality movie summaries, where both methods verified the appropriateness of the proposed methods. The annotation of the database and the code for the summarization system can be found at http://cognimuse.cs.ntua.gr/database.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Research related to computational modeling for machine-based understanding requires ground truth data for training, content analysis, and evaluation. In this paper, we present a multimodal video database, namely COGNIMUSE, annotated with sensory and semantic saliency, events, cross-media semantics, and emotion. The purpose of this database is manifold; it can be used for training and evaluation of event detection and summarization algorithms, for classification and recognition of audio-visual and cross-media events, as well as for emotion tracking. In order to enable comparisons with other computational models, we propose state-of-the-art algorithms, specifically a unified energy-based audio-visual framework and a method for text saliency computation, for the detection of perceptually salient events from videos. Additionally, a movie summarization system for the automatic production of summaries is presented. Two kinds of evaluation were performed, an objective based on the saliency annotation of the database and an extensive qualitative human evaluation of the automatically produced summaries, where we investigated what composes high-quality movie summaries, where both methods verified the appropriateness of the proposed methods. The annotation of the database and the code for the summarization system can be found at http://cognimuse.cs.ntua.gr/database. |
A Zlatintsi, I Rodomagoulakis, V Pitsikalis, P Koutras, N Kardaris, X Papageorgiou, C Tzafestas, P Maragos Social Human-Robot Interaction for the Elderly: Two Real-life Use Cases, Conference ACM/IEEE International Conference on Human-Robot Interaction (HRI), Vienna, Austria, 2017. Abstract | BibTeX | Links: [PDF] @conference{ZRP+17, title = {Social Human-Robot Interaction for the Elderly: Two Real-life Use Cases,}, author = {A Zlatintsi and I Rodomagoulakis and V Pitsikalis and P Koutras and N Kardaris and X Papageorgiou and C Tzafestas and P Maragos}, url = {http://robotics.ntua.gr/wp-content/publications/Zlatintsi+_SocialHRIforTheElderly_HRI-17.pdf}, year = {2017}, date = {2017-03-01}, booktitle = {ACM/IEEE International Conference on Human-Robot Interaction (HRI)}, address = {Vienna, Austria}, abstract = {We explore new aspects on assistive living via smart social human-robot interaction (HRI) involving automatic recognition of multimodal gestures and speech in a natural interface, providing social features in HRI. We discuss a whole framework of resources, including datasets and tools, briefly shown in two real-life use cases for elderly subjects: a multimodal interface of an assistive robotic rollator and an assistive bathing robot. We discuss these domain specific tasks, and open source tools, which can be used to build such HRI systems, as well as indicative results. Sharing such resources can open new perspectives in assistive HRI.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } We explore new aspects on assistive living via smart social human-robot interaction (HRI) involving automatic recognition of multimodal gestures and speech in a natural interface, providing social features in HRI. We discuss a whole framework of resources, including datasets and tools, briefly shown in two real-life use cases for elderly subjects: a multimodal interface of an assistive robotic rollator and an assistive bathing robot. We discuss these domain specific tasks, and open source tools, which can be used to build such HRI systems, as well as indicative results. Sharing such resources can open new perspectives in assistive HRI. |
G Karamanolakis, E Iosif, A Zlatintsi, A Pikrakis, A Potamianos Audio-based Distributional Semantic Models for Music Auto-tagging and Similarity Measurement Conference Proc. MultiLearn2017: Multimodal Processing, Modeling and Learning for Human-Computer/Robot Interaction Workshop, in conjuction with European Signal Processing Conference, Kos, Greece, 2017. Abstract | BibTeX | Links: [PDF] @conference{KIZ+17, title = {Audio-based Distributional Semantic Models for Music Auto-tagging and Similarity Measurement}, author = {G Karamanolakis and E Iosif and A Zlatintsi and A Pikrakis and A Potamianos}, url = {http://robotics.ntua.gr/wp-content/publications/Karamanolakis+_MultiLearn-17_ML7.pdf}, year = {2017}, date = {2017-01-01}, booktitle = {Proc. MultiLearn2017: Multimodal Processing, Modeling and Learning for Human-Computer/Robot Interaction Workshop, in conjuction with European Signal Processing Conference}, address = {Kos, Greece}, abstract = {The recent development of Audio-based Distributional Semantic Models (ADSMs) enables the computation of audio and lexical vector representations in a joint acoustic-semantic space. In this work, these joint representations are applied to the problem of automatic tag generation. The predicted tags together with their corresponding acoustic representation are exploited for the construction of acoustic-semantic clip embeddings. The proposed algorithms are evaluated on the task of similarity measurement between music clips. Acoustic-semantic models are shown to outperform the state-of-the-art for this task and produce high quality tags for audio/music clips.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } The recent development of Audio-based Distributional Semantic Models (ADSMs) enables the computation of audio and lexical vector representations in a joint acoustic-semantic space. In this work, these joint representations are applied to the problem of automatic tag generation. The predicted tags together with their corresponding acoustic representation are exploited for the construction of acoustic-semantic clip embeddings. The proposed algorithms are evaluated on the task of similarity measurement between music clips. Acoustic-semantic models are shown to outperform the state-of-the-art for this task and produce high quality tags for audio/music clips. |
2016 |
G Panagiotaropoulou, P Koutras, A Katsamanis, P Maragos, A Zlatintsi, A Protopapas, E Karavasilis, N Smyrnis fMRI-based Perceptual Validation of a computational Model for Visual and Auditory Saliency in Videos Conference Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing, Phoenix, AZ, USA, 2016. Abstract | BibTeX | Links: [PDF] @conference{PKK+16, title = {fMRI-based Perceptual Validation of a computational Model for Visual and Auditory Saliency in Videos}, author = {G Panagiotaropoulou and P Koutras and A Katsamanis and P Maragos and A Zlatintsi and A Protopapas and E Karavasilis and N Smyrnis}, url = {http://robotics.ntua.gr/wp-content/publications/PanagiotaropoulouEtAl_fMRI-Validation-CompAVsaliencyVideos_ICIP2016.pdf}, year = {2016}, date = {2016-09-01}, booktitle = {Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing}, address = {Phoenix, AZ, USA}, abstract = {In this study, we make use of brain activation data to investigate the perceptual plausibility of a visual and an auditory model for visual and auditory saliency in video processing. These models have already been successfully employed in a number of applications. In addition, we experiment with parameters, modifications and suitable fusion schemes. As part of this work, fMRI data from complex video stimuli were collected, on which we base our analysis and results. The core part of the analysis involves the use of well-established methods for the manipulation of fMRI data and the examination of variability across brain responses of different individuals. Our results indicate a success in confirming the value of these saliency models in terms of perceptual plausibility.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this study, we make use of brain activation data to investigate the perceptual plausibility of a visual and an auditory model for visual and auditory saliency in video processing. These models have already been successfully employed in a number of applications. In addition, we experiment with parameters, modifications and suitable fusion schemes. As part of this work, fMRI data from complex video stimuli were collected, on which we base our analysis and results. The core part of the analysis involves the use of well-established methods for the manipulation of fMRI data and the examination of variability across brain responses of different individuals. Our results indicate a success in confirming the value of these saliency models in terms of perceptual plausibility. |
G Karamanolakis, E Iosif, A Zlatintsi, A Pikrakis, A Potamianos Audio-Based Distributional Representations of Meaning Using a Fusion of Feature Encodings Conference 2016. Abstract | BibTeX | Links: [Webpage] [PDF] @conference{Karamanolakis2016, title = {Audio-Based Distributional Representations of Meaning Using a Fusion of Feature Encodings}, author = {G Karamanolakis and E Iosif and A Zlatintsi and A Pikrakis and A Potamianos}, url = {http://robotics.ntua.gr/wp-content/publications/Karamanolakis+_Interspeech16.PDF http://robotics.ntua.gr/wp-content/uploads/sites/2/karamanolakis16_interspeech.pdf}, year = {2016}, date = {2016-09-01}, abstract = {Recently a “Bag-of-Audio-Words” approach was proposed [1] for the combination of lexical features with audio clips in a multimodal semantic representation, i.e., an Audio Distributional Semantic Model (ADSM). An important step towards the creation of ADSMs is the estimation of the semantic distance between clips in the acoustic space, which is especially challenging given the diversity of audio collections. In this work, we investigate the use of different feature encodings in order to address this challenge following a two-step approach. First, an audio clip is categorized with respect to three classes, namely, music, speech and other. Next, the feature encodings are fused according to the posterior probabilities estimated in the previous step. Using a collection of audio clips annotated with tags we derive a mapping between words and audio clips. Based on this mapping and the proposed audio semantic distance, we construct an ADSM model in order to compute the distance between words (lexical semantic similarity task). The proposed model is shown to significantly outperform (23.6% relative improvement in correlation coefficient) the state-of-the-art results reported in the literature.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Recently a “Bag-of-Audio-Words” approach was proposed [1] for the combination of lexical features with audio clips in a multimodal semantic representation, i.e., an Audio Distributional Semantic Model (ADSM). An important step towards the creation of ADSMs is the estimation of the semantic distance between clips in the acoustic space, which is especially challenging given the diversity of audio collections. In this work, we investigate the use of different feature encodings in order to address this challenge following a two-step approach. First, an audio clip is categorized with respect to three classes, namely, music, speech and other. Next, the feature encodings are fused according to the posterior probabilities estimated in the previous step. Using a collection of audio clips annotated with tags we derive a mapping between words and audio clips. Based on this mapping and the proposed audio semantic distance, we construct an ADSM model in order to compute the distance between words (lexical semantic similarity task). The proposed model is shown to significantly outperform (23.6% relative improvement in correlation coefficient) the state-of-the-art results reported in the literature. |
Georgia Panagiotaropoulou, Petros Koutras, Athanasios Katsamanis, Petros Maragos, Athanasia Zlatintsi, Athanassios Protopapas, Efstratios Karavasilis, Nikolaos Smyrnis FMRI-based perceptual validation of a computational model for visual and auditory saliency in videos Conference Proceedings - International Conference on Image Processing, ICIP, 2016-August , 2016, ISSN: 15224880. Abstract | BibTeX | Links: [PDF] @conference{332, title = {FMRI-based perceptual validation of a computational model for visual and auditory saliency in videos}, author = { Georgia Panagiotaropoulou and Petros Koutras and Athanasios Katsamanis and Petros Maragos and Athanasia Zlatintsi and Athanassios Protopapas and Efstratios Karavasilis and Nikolaos Smyrnis}, url = {http://robotics.ntua.gr/wp-content/uploads/publications/PanagiotaropoulouEtAl_fMRI-Validation-CompAVsaliencyVideos_ICIP2016.pdf}, doi = {10.1109/ICIP.2016.7532447}, issn = {15224880}, year = {2016}, date = {2016-01-01}, booktitle = {Proceedings - International Conference on Image Processing, ICIP}, volume = {2016-August}, pages = {699--703}, abstract = {textcopyright 2016 IEEE.In this study, we make use of brain activation data to investigate the perceptual plausibility of a visual and an auditory model for visual and auditory saliency in video processing. These models have already been successfully employed in a number of applications. In addition, we experiment with parameters, modifications and suitable fusion schemes. As part of this work, fMRI data from complex video stimuli were collected, on which we base our analysis and results. The core part of the analysis involves the use of well-established methods for the manipulation of fMRI data and the examination of variability across brain responses of different individuals. Our results indicate a success in confirming the value of these saliency models in terms of perceptual plausibility.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } textcopyright 2016 IEEE.In this study, we make use of brain activation data to investigate the perceptual plausibility of a visual and an auditory model for visual and auditory saliency in video processing. These models have already been successfully employed in a number of applications. In addition, we experiment with parameters, modifications and suitable fusion schemes. As part of this work, fMRI data from complex video stimuli were collected, on which we base our analysis and results. The core part of the analysis involves the use of well-established methods for the manipulation of fMRI data and the examination of variability across brain responses of different individuals. Our results indicate a success in confirming the value of these saliency models in terms of perceptual plausibility. |
2015 |
A Zlatintsi, E.Iosif, P Maragos, A Potamianos Audio Salient Event Detection and Summarization using Audio and Text Modalities Conference Nice, France, 2015. Abstract | BibTeX | Links: [PDF] @conference{ZIM+15, title = {Audio Salient Event Detection and Summarization using Audio and Text Modalities}, author = {A Zlatintsi and E.Iosif and P Maragos and A Potamianos}, url = {http://robotics.ntua.gr/wp-content/publications/ZlatintsiEtAl_AudioTextSum-EUSIPCO-2015.pdf}, year = {2015}, date = {2015-09-01}, address = {Nice, France}, abstract = {This paper investigates the problem of audio event detection and summarization, building on previous work [1, 2] on the detection of perceptually important audio events based on saliency models. We take a synergistic approach to audio summarization where saliency computation of audio streams is assisted by using the text modality as well. Auditory saliency is assessed by auditory and perceptual cues such as Teager energy, loudness and roughness; all known to correlate with attention and human hearing. Text analysis incorporates part-of-speech tagging and affective modeling. A computational method for the automatic correction of the boundaries of the selected audio events is applied creating summaries that consist not only of salient but also meaningful and semantically coherent events. A non-parametric classification technique is employed and results are reported on the MovSum movie database using objective evaluations against ground-truth designating the auditory and semantically salient events.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } This paper investigates the problem of audio event detection and summarization, building on previous work [1, 2] on the detection of perceptually important audio events based on saliency models. We take a synergistic approach to audio summarization where saliency computation of audio streams is assisted by using the text modality as well. Auditory saliency is assessed by auditory and perceptual cues such as Teager energy, loudness and roughness; all known to correlate with attention and human hearing. Text analysis incorporates part-of-speech tagging and affective modeling. A computational method for the automatic correction of the boundaries of the selected audio events is applied creating summaries that consist not only of salient but also meaningful and semantically coherent events. A non-parametric classification technique is employed and results are reported on the MovSum movie database using objective evaluations against ground-truth designating the auditory and semantically salient events. |
P Koutras, A Zlatintsi, E.Iosif, A Katsamanis, P Maragos, A Potamianos Predicting Audio-Visual Salient Events Based on Visual, Audio and Text Modalities for Movie Summarization Conference Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing, Quebec, Canada, 2015. Abstract | BibTeX | Links: [PDF] @conference{KZI+15, title = {Predicting Audio-Visual Salient Events Based on Visual, Audio and Text Modalities for Movie Summarization}, author = {P Koutras and A Zlatintsi and E.Iosif and A Katsamanis and P Maragos and A Potamianos}, url = {http://robotics.ntua.gr/wp-content/publications/KZIKMP_MovieSum2_ICIP-2015.pdf}, year = {2015}, date = {2015-09-01}, booktitle = {Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing}, address = {Quebec, Canada}, abstract = {In this paper, we present a new and improved synergistic approach to the problem of audio-visual salient event detection and movie summarization based on visual, audio and text modalities. Spatio-temporal visual saliency is estimated through a perceptually inspired frontend based on 3D (space, time) Gabor filters and frame-wise features are extracted from the saliency volumes. For the auditory salient event detection we extract features based on Teager-Kaiser Energy Operator, while text analysis incorporates part-of-speech tag-ging and affective modeling of single words on the movie subtitles. For the evaluation of the proposed system, we employ an elementary and non-parametric classification technique like KNN. Detection results are reported on the MovSum database, using objective evaluations against ground-truth denoting the perceptually salient events, and human evaluations of the movie summaries. Our evaluation verifies the appropriateness of the proposed methods compared to our baseline system. Finally, our newly proposed summarization algorithm produces summaries that consist of salient and meaningful events, also improving the comprehension of the semantics.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this paper, we present a new and improved synergistic approach to the problem of audio-visual salient event detection and movie summarization based on visual, audio and text modalities. Spatio-temporal visual saliency is estimated through a perceptually inspired frontend based on 3D (space, time) Gabor filters and frame-wise features are extracted from the saliency volumes. For the auditory salient event detection we extract features based on Teager-Kaiser Energy Operator, while text analysis incorporates part-of-speech tag-ging and affective modeling of single words on the movie subtitles. For the evaluation of the proposed system, we employ an elementary and non-parametric classification technique like KNN. Detection results are reported on the MovSum database, using objective evaluations against ground-truth denoting the perceptually salient events, and human evaluations of the movie summaries. Our evaluation verifies the appropriateness of the proposed methods compared to our baseline system. Finally, our newly proposed summarization algorithm produces summaries that consist of salient and meaningful events, also improving the comprehension of the semantics. |
A Zlatintsi, P Koutras, N Efthymiou, P Maragos, A Potamianos, K Pastra Quality Evaluation of Computational Models for Movie Summarization Conference Costa Navarino, Messinia, Greece, 2015. Abstract | BibTeX | Links: [PDF] @conference{ZKE+15, title = {Quality Evaluation of Computational Models for Movie Summarization}, author = {A Zlatintsi and P Koutras and N Efthymiou and P Maragos and A Potamianos and K Pastra}, url = {http://robotics.ntua.gr/wp-content/publications/ZlatintsiEtAl_MovieSumEval-QoMEX2015.pdf}, year = {2015}, date = {2015-05-01}, address = {Costa Navarino, Messinia, Greece}, abstract = {In this paper we present a movie summarization system and we investigate what composes high quality movie summaries in terms of user experience evaluation. We propose state-of-the-art audio, visual and text techniques for the detection of perceptually salient events from movies. The evaluation of such computational models is usually based on the comparison of the similarity between the system-detected events and some ground-truth data. For this reason, we have developed the MovSum movie database, which includes sensory and semantic saliency annotation as well as cross-media relations, for objective evaluations. The automatically produced movie summaries were qualitatively evaluated, in an extensive human evaluation, in terms of informativeness and enjoyability accomplishing very high ratings up to 80% and 90%, respectively, which verifies the appropriateness of the proposed methods.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this paper we present a movie summarization system and we investigate what composes high quality movie summaries in terms of user experience evaluation. We propose state-of-the-art audio, visual and text techniques for the detection of perceptually salient events from movies. The evaluation of such computational models is usually based on the comparison of the similarity between the system-detected events and some ground-truth data. For this reason, we have developed the MovSum movie database, which includes sensory and semantic saliency annotation as well as cross-media relations, for objective evaluations. The automatically produced movie summaries were qualitatively evaluated, in an extensive human evaluation, in terms of informativeness and enjoyability accomplishing very high ratings up to 80% and 90%, respectively, which verifies the appropriateness of the proposed methods. |
P. Koutras, A. Zlatintsi, E. Iosif, A. Katsamanis, P. Maragos, A. Potamianos Predicting audio-visual salient events based on visual, audio and text modalities for movie summarization Conference Proceedings - International Conference on Image Processing, ICIP, 2015-December , 2015, ISSN: 15224880. @conference{307, title = {Predicting audio-visual salient events based on visual, audio and text modalities for movie summarization}, author = { P. Koutras and A. Zlatintsi and E. Iosif and A. Katsamanis and P. Maragos and A. Potamianos}, url = {http://robotics.ntua.gr/wp-content/uploads/publications/KZIKMP_MovieSum2_ICIP-2015.pdf}, doi = {10.1109/ICIP.2015.7351630}, issn = {15224880}, year = {2015}, date = {2015-01-01}, booktitle = {Proceedings - International Conference on Image Processing, ICIP}, volume = {2015-December}, pages = {4361--4365}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
2014 |
A Zlatintsi, P Maragos Comparison of Different Representations Based on Nonlinear Features for Music Genre Classification Conference Proc. European Signal Processing Conference, Lisbon, Portugal, 2014. Abstract | BibTeX | Links: [PDF] @conference{ZlMa14, title = {Comparison of Different Representations Based on Nonlinear Features for Music Genre Classification}, author = {A Zlatintsi and P Maragos}, url = {http://robotics.ntua.gr/wp-content/publications/ZlatintsiMaragos_MGC_EUSIPCO14_Lisbon_proc.pdf}, year = {2014}, date = {2014-09-01}, booktitle = {Proc. European Signal Processing Conference}, address = {Lisbon, Portugal}, abstract = {In this paper, we examine the descriptiveness and recognition properties of different feature representations for the analysis of musical signals, aiming in the exploration of their micro- and macro-structures, for the task of music genre classification. We explore nonlinear methods, such as the AM-FM model and ideas from fractal theory, so as to model the time-varying harmonic structure of musical signals and the geometrical complexity of the music waveform. The different feature representations’ efficacy is compared regarding their recognition properties for the specific task. The proposed features are evaluated against and in combination with Mel frequency cepstral coefficients (MFCC), using both static and dynamic classifiers, accomplishing an error reduction of 28%, illustrating that they can capture important aspects of music.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this paper, we examine the descriptiveness and recognition properties of different feature representations for the analysis of musical signals, aiming in the exploration of their micro- and macro-structures, for the task of music genre classification. We explore nonlinear methods, such as the AM-FM model and ideas from fractal theory, so as to model the time-varying harmonic structure of musical signals and the geometrical complexity of the music waveform. The different feature representations’ efficacy is compared regarding their recognition properties for the specific task. The proposed features are evaluated against and in combination with Mel frequency cepstral coefficients (MFCC), using both static and dynamic classifiers, accomplishing an error reduction of 28%, illustrating that they can capture important aspects of music. |
2013 |
Georgios Evangelopoulos, Athanasia Zlatintsi, Alexandros Potamianos, Petros Maragos, Konstantinos Rapantzikos, Georgios Skoumas, Yannis Avrithis Multimodal saliency and fusion for movie summarization based on aural, visual, and textual attention Journal Article IEEE Transactions on Multimedia, 15 (7), pp. 1553–1568, 2013, ISSN: 15209210. Abstract | BibTeX | Links: [PDF] @article{141, title = {Multimodal saliency and fusion for movie summarization based on aural, visual, and textual attention}, author = {Georgios Evangelopoulos and Athanasia Zlatintsi and Alexandros Potamianos and Petros Maragos and Konstantinos Rapantzikos and Georgios Skoumas and Yannis Avrithis}, url = {http://robotics.ntua.gr/wp-content/uploads/publications/EZPMRSA_MultimodalSaliencyFusionMovieSumAVTattention_ieeetMM13.pdf}, doi = {10.1109/TMM.2013.2267205}, issn = {15209210}, year = {2013}, date = {2013-01-01}, journal = {IEEE Transactions on Multimedia}, volume = {15}, number = {7}, pages = {1553--1568}, abstract = {Multimodal streams of sensory information are naturally parsed and integrated by humans using signal-level feature extraction and higher level cognitive processes. Detection of attention-invoking audiovisual segments is formulated in this work on the basis of saliency models for the audio, visual, and textual information conveyed in a video stream. Aural or auditory saliency is assessed by cues that quantify multifrequency waveform modulations, extracted through nonlinear operators and energy tracking. Visual saliency is measured through a spatiotemporal attention model driven by intensity, color, and orientation. Textual or linguistic saliency is extracted from part-of-speech tagging on the subtitles information available with most movie distributions. The individual saliency streams, obtained from modality-depended cues, are integrated in a multimodal saliency curve, modeling the time-varying perceptual importance of the composite video stream and signifying prevailing sensory events. The multimodal saliency representation forms the basis of a generic, bottom-up video summarization algorithm. Different fusion schemes are evaluated on a movie database of multimodal saliency annotations with comparative results provided across modalities. The produced summaries, based on low-level features and content-independent fusion and selection, are of subjectively high aesthetic and informative quality.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Multimodal streams of sensory information are naturally parsed and integrated by humans using signal-level feature extraction and higher level cognitive processes. Detection of attention-invoking audiovisual segments is formulated in this work on the basis of saliency models for the audio, visual, and textual information conveyed in a video stream. Aural or auditory saliency is assessed by cues that quantify multifrequency waveform modulations, extracted through nonlinear operators and energy tracking. Visual saliency is measured through a spatiotemporal attention model driven by intensity, color, and orientation. Textual or linguistic saliency is extracted from part-of-speech tagging on the subtitles information available with most movie distributions. The individual saliency streams, obtained from modality-depended cues, are integrated in a multimodal saliency curve, modeling the time-varying perceptual importance of the composite video stream and signifying prevailing sensory events. The multimodal saliency representation forms the basis of a generic, bottom-up video summarization algorithm. Different fusion schemes are evaluated on a movie database of multimodal saliency annotations with comparative results provided across modalities. The produced summaries, based on low-level features and content-independent fusion and selection, are of subjectively high aesthetic and informative quality. |
A Zlatintsi Music Signal Processing and Applications in Recognition PhD Thesis School of ECE, NTUA, 2013. Abstract | BibTeX | Links: [PDF] @phdthesis{Zlatintsi13, title = {Music Signal Processing and Applications in Recognition}, author = {A Zlatintsi}, url = {http://robotics.ntua.gr/wp-content/publications/Zlatintsi_PhDThesis_Dec2013_EMP.pdf}, year = {2013}, date = {2013-12-01}, school = {School of ECE, NTUA}, abstract = {This thesis lays in the area of signal processing and analysis of music signalsusing computational methods for the extraction of effective representations for automatic recognition. We explore and develop efficient algorithms using nonlinear methods for the analysis of the structure of music signals, which is of importance for their modeling. Our main research directions deals with the analysis of the structure and the characteristics of musical instruments in order to gain insight about their function and properties. We study the characteristics of the different genres of music.Finally, we evaluate the effectiveness of the proposed nonlinear models for the detection of perceptually important music and audio events. The approach we follow contributes to state-of-the-art technologies related to automatic computer-based recognition of musical signals and audio summarization, which nowadays are essential in everyday life. Because of the vast amount of music, audio and multimedia data in the web and our personal computers, the use of this study could be shown in applications such as automatic genre classification, automatic recognition of music’s basic structures, such as musical instruments, and audio content analysis for music and audio summarization. The above mentioned applications require robust solutions to information processing problems. Toward this goal, the development of efficient digital signal processing methods and the extraction of relevant features is of importance. In this thesis we propose such methods and algorithms for feature extraction with interesting results that render the descriptors of direct applicability. The proposed methods are applied on classification experiments illustrating that they can capture important aspects of music, such as the micro-variations of their structure. Descriptors based on macro-structures may reduce the complexity of the classification system, since satisfactory results can be achieved using simpler statistical models. Finally, the introduction of a ‘‘music’’ filterbank appears to be promising for automatic genre classification.}, keywords = {}, pubstate = {published}, tppubtype = {phdthesis} } This thesis lays in the area of signal processing and analysis of music signalsusing computational methods for the extraction of effective representations for automatic recognition. We explore and develop efficient algorithms using nonlinear methods for the analysis of the structure of music signals, which is of importance for their modeling. Our main research directions deals with the analysis of the structure and the characteristics of musical instruments in order to gain insight about their function and properties. We study the characteristics of the different genres of music.Finally, we evaluate the effectiveness of the proposed nonlinear models for the detection of perceptually important music and audio events. The approach we follow contributes to state-of-the-art technologies related to automatic computer-based recognition of musical signals and audio summarization, which nowadays are essential in everyday life. Because of the vast amount of music, audio and multimedia data in the web and our personal computers, the use of this study could be shown in applications such as automatic genre classification, automatic recognition of music’s basic structures, such as musical instruments, and audio content analysis for music and audio summarization. The above mentioned applications require robust solutions to information processing problems. Toward this goal, the development of efficient digital signal processing methods and the extraction of relevant features is of importance. In this thesis we propose such methods and algorithms for feature extraction with interesting results that render the descriptors of direct applicability. The proposed methods are applied on classification experiments illustrating that they can capture important aspects of music, such as the micro-variations of their structure. Descriptors based on macro-structures may reduce the complexity of the classification system, since satisfactory results can be achieved using simpler statistical models. Finally, the introduction of a ‘‘music’’ filterbank appears to be promising for automatic genre classification. |
2012 |
A Zlatintsi, P Maragos, A Potamianos, G Evangelopoulos A Saliency-Based Approach to Audio Event Detection and Summarization Conference Proc. European Signal Processing Conference, Bucharest, Romania, 2012. Abstract | BibTeX | Links: [PDF] @conference{ZMP+12, title = {A Saliency-Based Approach to Audio Event Detection and Summarization}, author = {A Zlatintsi and P Maragos and A Potamianos and G Evangelopoulos}, url = {http://robotics.ntua.gr/wp-content/publications/ZlatintsiMaragos+_SaliencyBasedAudioSummarization_EUSIPCO2012.pdf}, year = {2012}, date = {2012-08-01}, booktitle = {Proc. European Signal Processing Conference}, address = {Bucharest, Romania}, abstract = {In this paper, we approach the problem of audio summarization by saliency computation of audio streams, exploring the potential of a modulation model for the detection of perceptually important audio events based on saliency models, along with various fusion schemes for their combination. The fusion schemes include linear, adaptive and nonlinear methods. A machine learning approach, where training of the features is performed, was also applied for the purpose of comparison with the proposed technique. For the evaluation of the algorithm we use audio data taken from movies and we show that nonlinear fusion schemes perform best. The results are reported on the MovSum database, using objective evaluations (against ground-truth denoting the perceptually important audio events). Analysis of the selected audio segments is also performed against a labeled database in respect to audio categories, while a method for fine-tuning of the selected audio events is proposed.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this paper, we approach the problem of audio summarization by saliency computation of audio streams, exploring the potential of a modulation model for the detection of perceptually important audio events based on saliency models, along with various fusion schemes for their combination. The fusion schemes include linear, adaptive and nonlinear methods. A machine learning approach, where training of the features is performed, was also applied for the purpose of comparison with the proposed technique. For the evaluation of the algorithm we use audio data taken from movies and we show that nonlinear fusion schemes perform best. The results are reported on the MovSum database, using objective evaluations (against ground-truth denoting the perceptually important audio events). Analysis of the selected audio segments is also performed against a labeled database in respect to audio categories, while a method for fine-tuning of the selected audio events is proposed. |
A Zlatintsi, P Maragos AM-FM Modulation Features for Music Instrument Signal Analysis and Recognition Conference Proc. European Signal Processing Conference, Bucharest, Romania, 2012. Abstract | BibTeX | Links: [PDF] @conference{ZlMa12, title = {AM-FM Modulation Features for Music Instrument Signal Analysis and Recognition}, author = {A Zlatintsi and P Maragos}, url = {http://robotics.ntua.gr/wp-content/publications/ZlatintsiMaragos_MusicalInstrumentsAMFM_EUSIPCO2012.pdf}, year = {2012}, date = {2012-08-01}, booktitle = {Proc. European Signal Processing Conference}, address = {Bucharest, Romania}, abstract = {In this paper, we explore a nonlinear AM-FM model to extract alternative features for music instrument recognition tasks. Amplitude and frequency micro-modulations are measured in musical signals and are employed to model the existing information. The features used are the multiband mean instantaneous amplitude (mean-IAM) and mean instantaneous frequency (mean-IFM) modulation. The instantaneous features are estimated using the multiband Gabor Energy Separation Algorithm (Gabor-ESA). An alternative method, the iterative-ESA is also explored; and initial experimentation shows that it could be used to estimate the harmonic content of a tone. The Gabor-ESA is evaluated against and in combination with Mel frequency cepstrum coefficients (MFCCs) using both static and dynamic classifiers. The method used in this paper has proven to be able to extract the fine-structured modulations of music signals; further, it has shown to be promising for recognition tasks accomplishing an error rate reduction up to 60% for the best recognition case combined with MFCCs.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this paper, we explore a nonlinear AM-FM model to extract alternative features for music instrument recognition tasks. Amplitude and frequency micro-modulations are measured in musical signals and are employed to model the existing information. The features used are the multiband mean instantaneous amplitude (mean-IAM) and mean instantaneous frequency (mean-IFM) modulation. The instantaneous features are estimated using the multiband Gabor Energy Separation Algorithm (Gabor-ESA). An alternative method, the iterative-ESA is also explored; and initial experimentation shows that it could be used to estimate the harmonic content of a tone. The Gabor-ESA is evaluated against and in combination with Mel frequency cepstrum coefficients (MFCCs) using both static and dynamic classifiers. The method used in this paper has proven to be able to extract the fine-structured modulations of music signals; further, it has shown to be promising for recognition tasks accomplishing an error rate reduction up to 60% for the best recognition case combined with MFCCs. |
2011 |
A Zlatintsi, P Maragos Musical Instruments Signal Analysis and Recognition Using Fractal Features Conference Proc. European Signal Processing Conference, Barcelona, Spain, 2011. Abstract | BibTeX | Links: [PDF] @conference{ZlMa11, title = {Musical Instruments Signal Analysis and Recognition Using Fractal Features}, author = {A Zlatintsi and P Maragos}, url = {http://robotics.ntua.gr/wp-content/publications/ZlatintsiMaragos_MusicalInstrumentsMFD_EUSIPCO2011.pdf}, year = {2011}, date = {2011-08-01}, booktitle = {Proc. European Signal Processing Conference}, address = {Barcelona, Spain}, abstract = {Analyzing the structure of music signals at multiple time scales is of importance both for modeling music signals and their automatic computer-based recognition. In this paper we propose the multiscale fractal dimension prourl as a descriptor useful to quantify the multiscale complexity of the music waveform. We have experimentally found that this descriptor can discriminate several aspects among different music instruments. We compare the descriptiveness of our features against that of Mel frequency cepstral coefficients (MFCCs) using both static and dynamic classifiers, such as Gaussian mixture models (GMMs) and hidden Markov models (HMMs). The methods and features proposed in this paper are promising for music signal analysis and of direct applicability in large-scale music classification tasks.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Analyzing the structure of music signals at multiple time scales is of importance both for modeling music signals and their automatic computer-based recognition. In this paper we propose the multiscale fractal dimension prourl as a descriptor useful to quantify the multiscale complexity of the music waveform. We have experimentally found that this descriptor can discriminate several aspects among different music instruments. We compare the descriptiveness of our features against that of Mel frequency cepstral coefficients (MFCCs) using both static and dynamic classifiers, such as Gaussian mixture models (GMMs) and hidden Markov models (HMMs). The methods and features proposed in this paper are promising for music signal analysis and of direct applicability in large-scale music classification tasks. |
N Malandrakis, A Potamianos, G Evangelopoulos, A Zlatintsi A Supervised Approach to Movie Emotion Tracking Conference Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing, Prague, Czech Republic, 2011. Abstract | BibTeX | Links: [PDF] @conference{MPE+11, title = {A Supervised Approach to Movie Emotion Tracking}, author = {N Malandrakis and A Potamianos and G Evangelopoulos and A Zlatintsi}, url = {http://robotics.ntua.gr/wp-content/publications/Malandrakis+_movie_emotion_ICASSP11.pdf}, year = {2011}, date = {2011-05-01}, booktitle = {Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing}, address = {Prague, Czech Republic}, abstract = {In this paper, we present experiments on continuous time, continuous scale affective movie content recognition (emotion tracking). A major obstacle for emotion research has been the lack of appropriately annotated databases, limiting the potential for supervised algorithms. To that end we develop and present a database of movie affect, annotated in continuous time, on a continuous valence-arousal scale. Supervised learning methods are proposed to model the continuous affective response using hidden Markov Models (independent) in each dimension. These models classify each video frame into one of seven discrete categories (in each dimension); the discrete-valued curves are then converted to continuous values via spline interpolation. A variety of audio-visual features are investigated and an optimal feature set is selected. The potential of the method is experimentally verified on twelve 30-minute movie clips with good precision at a macroscopic level.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } In this paper, we present experiments on continuous time, continuous scale affective movie content recognition (emotion tracking). A major obstacle for emotion research has been the lack of appropriately annotated databases, limiting the potential for supervised algorithms. To that end we develop and present a database of movie affect, annotated in continuous time, on a continuous valence-arousal scale. Supervised learning methods are proposed to model the continuous affective response using hidden Markov Models (independent) in each dimension. These models classify each video frame into one of seven discrete categories (in each dimension); the discrete-valued curves are then converted to continuous values via spline interpolation. A variety of audio-visual features are investigated and an optimal feature set is selected. The potential of the method is experimentally verified on twelve 30-minute movie clips with good precision at a macroscopic level. |
Athanasia Zlatintsi, Petros Maragos MUSICAL INSTRUMENTS SIGNAL ANALYSIS AND RECOGNITION USING FRACTAL FEATURES Conference Proc. 19th European Signal Processing Conference (EUSIPCO-2011), (Eusipco), 2011. @conference{182, title = {MUSICAL INSTRUMENTS SIGNAL ANALYSIS AND RECOGNITION USING FRACTAL FEATURES}, author = { Athanasia Zlatintsi and Petros Maragos}, url = {http://robotics.ntua.gr/wp-content/uploads/publications/ZlatintsiMaragos_MusicalInstrumentsMFD_EUSIPCO2011.pdf}, year = {2011}, date = {2011-01-01}, booktitle = {Proc. 19th European Signal Processing Conference (EUSIPCO-2011)}, number = {Eusipco}, pages = {684--688}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
2009 |
G Evangelopoulos, A Zlatintsi, G Skoumas, K Rapantzikos, A Potamianos, P Maragos, Y Avrithis Video Event Detection and Summarization Using Audio, Visual and Text Saliency Conference Taipei, Taiwan, 2009. Abstract | BibTeX | Links: [PDF] @conference{EZS+09, title = {Video Event Detection and Summarization Using Audio, Visual and Text Saliency}, author = {G Evangelopoulos and A Zlatintsi and G Skoumas and K Rapantzikos and A Potamianos and P Maragos and Y Avrithis}, url = {http://robotics.ntua.gr/wp-content/publications/EvangelopoulosZlatintsiEtAl_VideoEventDetectionSummarizationUsingAVTSaliency_ICASSP09.pdf}, year = {2009}, date = {2009-04-01}, address = {Taipei, Taiwan}, abstract = {Detection of perceptually important video events is formulated here on the basis of saliency models for the audio, visual and textual information conveyed in a video stream. Audio saliency is assessed by cues that quantify multifrequency waveform modulations, extracted through nonlinear operators and energy tracking. Visual saliency is measured through a spatiotemporal attention model driven by intensity, color and motion. Text saliency is extracted from part-of-speech tagging on the subtitles information available with most movie distributions. The various modality curves are integrated in a single attention curve, where the presence of an event may be signised in one or multiple domains. This multimodal saliency curve is the basis of a bottom-up video summarization algorithm, that refines results from unimodal or audiovisual-based skimming. The algorithm performs favorably for video summarization in terms of informativeness and enjoyability.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Detection of perceptually important video events is formulated here on the basis of saliency models for the audio, visual and textual information conveyed in a video stream. Audio saliency is assessed by cues that quantify multifrequency waveform modulations, extracted through nonlinear operators and energy tracking. Visual saliency is measured through a spatiotemporal attention model driven by intensity, color and motion. Text saliency is extracted from part-of-speech tagging on the subtitles information available with most movie distributions. The various modality curves are integrated in a single attention curve, where the presence of an event may be signised in one or multiple domains. This multimodal saliency curve is the basis of a bottom-up video summarization algorithm, that refines results from unimodal or audiovisual-based skimming. The algorithm performs favorably for video summarization in terms of informativeness and enjoyability. |
G Evangelopoulos, A Zlatintsi, G Skoumas, K Rapantzikos, A Potamianos, P Maragos, Y Avrithis Video Event Detection and Summarization using Audio, Visual and Text Saliency Conference Icassp, (2), 2009, ISBN: 9781424423545. @conference{195, title = {Video Event Detection and Summarization using Audio, Visual and Text Saliency}, author = { G Evangelopoulos and A Zlatintsi and G Skoumas and K Rapantzikos and A Potamianos and P Maragos and Y Avrithis}, url = {http://robotics.ntua.gr/wp-content/uploads/publications/EvangelopoulosZlatintsiEtAl_VideoEventDetectionSummarizationUsingAVTSaliency_ICASSP09.pdf}, isbn = {9781424423545}, year = {2009}, date = {2009-01-01}, booktitle = {Icassp}, number = {2}, pages = {3553--3556}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
2008 |
G Evangelopoulos, K Rapantzikos, A Potamianos, P Maragos, A Zlatintsi, Y Avrithis Movie Summarization based on Audiovisual Saliency Detection Conference Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing, San Diego, CA, U.S.A., 2008. Abstract | BibTeX | Links: [PDF] @conference{ERP+08, title = {Movie Summarization based on Audiovisual Saliency Detection}, author = {G Evangelopoulos and K Rapantzikos and A Potamianos and P Maragos and A Zlatintsi and Y Avrithis}, url = {http://robotics.ntua.gr/wp-content/publications/EvangelopoulosRapantzikosEtAl_MovieSum_ICIP2008_fancyhead.pdf}, year = {2008}, date = {2008-10-01}, booktitle = {Proc. {IEEE} Int'l Conf. Acous., Speech, and Signal Processing}, address = {San Diego, CA, U.S.A.}, abstract = {Based on perceptual and computational attention modeling studies, we formulate measures of saliency for an audiovisual stream. Audio saliency is captured by signal modulations and related multi-frequency band features, extracted through nonlinear operators and energy tracking. Visual saliency is measured by means of a spatiotemporal attention model driven by various feature cues (intensity, color, motion). Audio and video curves are integrated in a single attention curve, where events may be enhanced, suppressed or vanished. The presence of salient events is signified on this audiovisual curve by geometrical features such as local extrema, sharp transition points and level sets. An audiovisual saliency-based movie summarization algorithm is proposed and evaluated. The algorithm is shown to perform very well in terms of summary informativeness and enjoyability for movie clips of various genres.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Based on perceptual and computational attention modeling studies, we formulate measures of saliency for an audiovisual stream. Audio saliency is captured by signal modulations and related multi-frequency band features, extracted through nonlinear operators and energy tracking. Visual saliency is measured by means of a spatiotemporal attention model driven by various feature cues (intensity, color, motion). Audio and video curves are integrated in a single attention curve, where events may be enhanced, suppressed or vanished. The presence of salient events is signified on this audiovisual curve by geometrical features such as local extrema, sharp transition points and level sets. An audiovisual saliency-based movie summarization algorithm is proposed and evaluated. The algorithm is shown to perform very well in terms of summary informativeness and enjoyability for movie clips of various genres. |
D Spachos, A Zlatintsi, V Moschou, P Antonopoulos, E Benetos, M Kotti, K Tzimouli, C Kotropoulos, N Nikolaidis, P Maragos, I Pitas MUSCLE Movie Database: A Multimodal Corpus With Rich Annotation For Dialogue And Saliency Detection Conference Marrakech, Morocco, 2008. Abstract | BibTeX | Links: [PDF] @conference{SZM+-8, title = {MUSCLE Movie Database: A Multimodal Corpus With Rich Annotation For Dialogue And Saliency Detection}, author = {D Spachos and A Zlatintsi and V Moschou and P Antonopoulos and E Benetos and M Kotti and K Tzimouli and C Kotropoulos and N Nikolaidis and P Maragos and I Pitas}, url = {http://robotics.ntua.gr/wp-content/publications/SpachosZlatintsi+_MuscleMovieDatabase_LREC08.pdf}, year = {2008}, date = {2008-05-01}, address = {Marrakech, Morocco}, abstract = {Semantic annotation of multimedia content is important for training, testing, and assessing content-based algorithms for indexing, organization, browsing, and retrieval. To this end, an annotated multimodal movie corpus has been collected to be used as a test bed for development and assessment of content-based multimedia processing, such as speaker clustering, speaker turn detection, visual speech activity detection, face detection, face clustering, scene segmentation, saliency detection, and visual dialogue detection. All metadata are saved in XML format following the MPEG-7 ISO prototype to ensure data compatibility and reusability. The entire MUSCLE movie database is available for download through the web. Visual speech activity and dialogue detection algorithms that have been developed within the software package DIVA3D and tested on this database are also briefly described. Furthermore, we review existing annotation tools with emphasis on the novel annotation tool Anthropos7 Editor.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Semantic annotation of multimedia content is important for training, testing, and assessing content-based algorithms for indexing, organization, browsing, and retrieval. To this end, an annotated multimodal movie corpus has been collected to be used as a test bed for development and assessment of content-based multimedia processing, such as speaker clustering, speaker turn detection, visual speech activity detection, face detection, face clustering, scene segmentation, saliency detection, and visual dialogue detection. All metadata are saved in XML format following the MPEG-7 ISO prototype to ensure data compatibility and reusability. The entire MUSCLE movie database is available for download through the web. Visual speech activity and dialogue detection algorithms that have been developed within the software package DIVA3D and tested on this database are also briefly described. Furthermore, we review existing annotation tools with emphasis on the novel annotation tool Anthropos7 Editor. |
D Spachos, A Zlatintsi MUSCLE movie database: A multimodal corpus with rich annotation for dialogue and saliency detection Conference Programme of the Workshop on Multimodal Corpora, 2008. @conference{211, title = {MUSCLE movie database: A multimodal corpus with rich annotation for dialogue and saliency detection}, author = { D Spachos and A Zlatintsi}, url = {http://users.uoi.gr/cs01702/MargaritaKotti/MypublicationsPDFs/Muscle movie.pdf}, year = {2008}, date = {2008-01-01}, booktitle = {Programme of the Workshop on Multimodal Corpora}, pages = {16}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
G. Evangelopoulos, K. Rapantzikos, A. Potamianos, P. Maragos, A. Zlatintsi, Y. Avrithis Movie summarization based on audiovisual saliency detection Conference Proceedings - International Conference on Image Processing, ICIP, 2008, ISSN: 15224880. Abstract | BibTeX | Links: [PDF] @conference{203, title = {Movie summarization based on audiovisual saliency detection}, author = { G. Evangelopoulos and K. Rapantzikos and A. Potamianos and P. Maragos and A. Zlatintsi and Y. Avrithis}, url = {http://robotics.ntua.gr/wp-content/uploads/publications/ERPMZA_MovieSummarizAVSaliency_ICIP2008.pdf}, doi = {10.1109/ICIP.2008.4712308}, issn = {15224880}, year = {2008}, date = {2008-01-01}, booktitle = {Proceedings - International Conference on Image Processing, ICIP}, pages = {2528--2531}, abstract = {Based on perceptual and computational attention modeling studies, we formulate measures of saliency for an audiovisual stream. Audio saliency is captured by signal modulations and related multi-frequency band features, extracted through nonlinear operators and energy tracking. Visual saliency is measured by means of a spatiotemporal attention model driven by various feature cues (intensity, color, motion). Audio and video curves are integrated in a single attention curve, where events may be enhanced, suppressed or vanished. The presence of salient events is signified on this audiovisual curve by geometrical features such as local extrema, sharp transition points and level sets. An audiovisual saliency-based movie summarization algorithm is proposed and evaluated. The algorithm is shown to perform very well in terms of summary informativeness and enjoyability for movie clips of various genres.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Based on perceptual and computational attention modeling studies, we formulate measures of saliency for an audiovisual stream. Audio saliency is captured by signal modulations and related multi-frequency band features, extracted through nonlinear operators and energy tracking. Visual saliency is measured by means of a spatiotemporal attention model driven by various feature cues (intensity, color, motion). Audio and video curves are integrated in a single attention curve, where events may be enhanced, suppressed or vanished. The presence of salient events is signified on this audiovisual curve by geometrical features such as local extrema, sharp transition points and level sets. An audiovisual saliency-based movie summarization algorithm is proposed and evaluated. The algorithm is shown to perform very well in terms of summary informativeness and enjoyability for movie clips of various genres. |