Éditeur(s) : HAL CCSD Institute of Electrical and Electronics Engineers
Résumé : International audience
This paper presents our response to the first interna- tional challenge on Facial Emotion Recognition and Analysis. We propose to combine different types of features to automatically detect Action Units in facial images. We use one multi-kernel SVM for each Action Unit we want to detect. The first kernel matrix is computed using Local Gabor Binary Pattern histograms and a histogram intersection kernel. The second kernel matrix is computed from AAM coefficients and an RBF kernel. During the training step, we combine these two types of features using the recently proposed SimpleMKL algorithm. SVM outputs are then averaged to exploit temporal information in the sequence. To eval- uate our system, we perform deep experimentations on several key issues: influence of features and kernel function in histogram- based SVM approaches, influence of spatially-independent in- formation versus geometric local appearance information and benefits of combining both, sensitivity to training data and interest of temporal context adaptation. We also compare our results to those of the other participants and try to explain why our method had the best performance during the FERA challenge.
ISSN: 1083-4419

hal-00731864
http://hal.upmc.fr/hal-00731864
http://hal.upmc.fr/hal-00731864/document
http://hal.upmc.fr/hal-00731864/file/senechal12tsmc_preprint.pdf
DOI : 10.1109/TSMCB.2012.2193567

Éditeur(s) : HAL CCSD IEEE Computer Society
Résumé : 5 pages, 2 figures, composed with Microsoft Word. Proceedings of the Seventh International Conference on Document Analysis and Recognition (ICDAR 2003). Edinburgh, Écosse, 3-6 août 2003. Également disponible à l'URL : http://www.cse.salford.ac.uk/prima/ICDAR2003/Papers/0100_563_likforman_l.pdf
International audience
In the frame of a Unified Messaging System, a crucial task of the system is to provide the user with key information on every message received, like keywords reflecting the object of the message, or the name of the sender. However, in the case of facsimiles, this information is not as easy to detect as in the case of e-mails, since no standard headers are defined. The aim of the present work is to identify and extract a specific information (the name of the sender) from a fax cover page. For this purpose, methods based on image document analysis (OCR recognition, physical blocks selection), and text analysis methods (optimised dictionary lookup, local grammar rules), are implemented to work in parallel. The fusion of their results brings a more accurate guess than any of the methods would achieve separately.
Proceedings of the Seventh International Conference on Document Analysis and Recognition (ICDAR 2003)
Edinburgh, United Kingdom

hal-00328998
https://hal.archives-ouvertes.fr/hal-00328998
DOI : 10.1109/ICDAR.2003.1227724

Éditeur(s) : HAL CCSD
Résumé : International audience
Proceedings of the 15th Conference of the International Graphonomics Society - IGS 2011
Cancun, Mexico

hal-00604378
https://hal.archives-ouvertes.fr/hal-00604378

Éditeur(s) : HAL CCSD
Résumé : International audience
L'avènement des interfaces orientées stylo à conduit au développement de nombreuses applications concernant l'analyse d'une forme dessinée à main levée. Mais un dessin, ce n'est pas uniquement une forme, c'est également l'ensemble des gestes constituant l'activité de tracé ayant donné naissance à la trace. Les interfaces orientées stylo nous donnent la possibilité d'analyser cette dimension du dessin. Nous appelons ce type d'analyse : analyse procédurale. Dans cet article, nous allons montrer comment identifier les procédures de dessin privilégiées des utilisateurs, et ce malgré une grande variabilité intra-scripteur, en utilisant une méthode de filtrage du bruit procédural. Puis nous montrerons un exemple d'application de cette approche en proposant une méthode de reconnaissance de dessinateur basée sur l'analyse procédurale.
Actes du Colloque International Francophone sur l'Écrit et le Document 2012
Bordeaux, France

hal-00698875
https://hal.archives-ouvertes.fr/hal-00698875

Résumé : Cette thèse traite de la problématique de l'analyse automatique de tracés manuscrits réalisés à mains levée, à caractère géométrique et acquis en ligne. Lorsque l'on analyse un tracé en ligne, on peut le faire selon plusieurs angles. Comme pour les tracés hors ligne, on peu s'atteler à reconnaitre la forme produite. Mais le caractère en ligne de l'acquisition permet également d'autres niveaux d'analyse, comme l'analyse de la procédure utilisée par le dessinateur lors de la production de son dessin. Nous avons dans le cadre de ce travail élaboré des outils permettant de telles analyses de la production de tracé. Dans une première partie nous traitons des pré traitements à effectuer sur le tracé afin de pouvoir procéder aux analyses spatio-graphique et procédurale. Ces pré traitements sont le filtrage, la segmentation mixte à base d'arcs de cercles et de segments de droite, la sélection des points d'intérêt ainsi que l'identification de la nature de ces points. La seconde partie concerne l'analyse visuo-spatiale. Nous abordons alors deux aspects: la question de la reconnaissance des formes et celle de l'évaluation automatique de leur qualité. Notre contribution consiste en une méthode de reconnaissance basée sur l'apparence utilisant des descripteurs locaux. La dernière partie traite de l'analyse procédurale. Nous proposons une méthode d'extraction et de modélisation de la procédure mise en œuvre lors du dessin d'une forme. Nous montrons comment en utilisant cette modélisation, nous pouvons non seulement déterminer la procédure favorite d'un utilisateur mais également reconnaitre un dessinateur par l'analyse de sa procédure.
This thesis handles the problem of the automatic analysis of online hand drawn geometric sketches. An online sketch can be analysed according to several points of view. As for offline sketching, we can try to recognize the produced shape. However, online sketching allows other levels of analysis. For example the analysis of the behavior of the drawer during the production of the sketch. In this thesis, we have tried to develop tools allowing a multi level analysis, including both shape and behavior analysis. The first part of our work deals with the pre treatments that must be performed on the sketch in order to allow upper level analysis. Those pre treatments are filtering, mixed segmentation and feature points detection and labelisation. In the second part, we approach shape analysis in two aspects: shape recognition and evaluation. We have developed a appearance based method which use local descriptors to allows both recognition and evaluation of the quality of a produced shape compared to the model . in the last part we propose a method for drawer's behavior extraction and modeling. Then we show how we can not only determine the favorite procedure of a drawer but also recognize the drawer by analyzing his behavior.

http://www.theses.fr/2010AGUY0394/document

Éditeur(s) : HAL CCSD Springer
Résumé : Third International Workshop, MLMI 2012, Held in Conjunction with MICCAI 2012
International audience
Universal Nearest Neighbours (unn) is a classifier recently proposed, which can also effectively estimates the posterior probability of each classification act. This algorithm, intrinsically binary, requires the use of a decomposition method to cope with multiclass problems, thus reducing their complexity in less complex binary subtasks. Then, a reconstruction rule provides the final classification. In this paper we show that the application of unn algorithm in conjunction with a reconstruction rule based on the posterior probabilities provides a classification scheme robust among different biomedical image datasets. To this aim, we compare unn performance with those achieved by Support Vector Machine with two different kernels and by a k Nearest Neighbours classifier, and applying two different reconstruction rules for each of the aforementioned classification paradigms. The results on one private and five public biomedical datasets show satisfactory performance.
Machine Learning in Medical Imaging
Nice, France

hal-00958909
https://hal.archives-ouvertes.fr/hal-00958909
https://hal.archives-ouvertes.fr/hal-00958909/document
https://hal.archives-ouvertes.fr/hal-00958909/file/Miccai.pdf
DOI : 10.1007/978-3-642-35428-1_15

Éditeur(s) : HAL CCSD
Résumé : Weather stations are constantly provided with thousands of satellite images with a view to anticipating the progress of natural phenomenons. This constant flow ofinformation requires speed and a strict analysis. Consequently, it is absolutely necessary to implement systems capable of working night and day, in order to make assessments. In this article, we suggest a method relying on pattern recognition investigation techniques, such as segmentation and classification. These techniques will be used to locate and discriminate the banks of clouds coming from images picked up by the GOES geostationary satellite. Treatment of the cloud shapes which have been found can be carried out, using operations, such as mathematical morphology, so as to draw discriminating characteristics. Classifier systems are used in order to interpret the characteristics which have been obtained and to conduct cloud classification.
15th Meeting of the Caribbean Academy of Sciences
Gosier, Guadeloupe

hal-00524903
https://hal.archives-ouvertes.fr/hal-00524903

Éditeur(s) : HAL CCSD
Résumé : Session "Articles"
National audience
Dans cet article, nous présentons une méthode robuste et précise pour détecter 17 points caractéristiques du visage sur des images expressives. Une nouvelle architecture multi-résolution basée sur les récents algorithmes multi-noyau est introduite. Les patches de faibles résolutions codent les informations globales du visage donnant lieu à une détection grossière mais robuste du point désiré. Les patches de grandes résolutions quant à eux utilisent les détails locaux afin d'affiner cette localisation. En combinant une détection indépendante de points et des informations a priori sur les distributions de points, nous proposons un détecteur robuste aux changements d'expressions ainsi qu'aux variations d'illuminations. Ce système a été évalué sur plusieurs bases de données de la littérature. Les résultats présentés améliorent les performances des détecteurs de l'état de l'art
Actes de la conférence RFIA 2012
Lyon, France

hal-00656522
https://hal.archives-ouvertes.fr/hal-00656522
https://hal.archives-ouvertes.fr/hal-00656522/document
https://hal.archives-ouvertes.fr/hal-00656522/file/rfia2012_submission_99.pdf

Éditeur(s) : HAL CCSD
Résumé : Session "Articles"
National audience
Nous présentons ici une méthode permettant de créer et de représenter sous forme vectorielle des classes de transition à partir d'une classification supervisée basée sur des arbres de décision flous. Cette méthode est particulièrement utile pour la classification stricte de données dont les frontières sont naturellement diffuses ainsi que pour leur représentation au sein d'un System d'Information Géographique. C'est le cas par exemple des formations forestières entre lesquels les transitions sont généralement étalées sur plusieurs centaines de mètres avec des gradients dépendant de condition environnementales locales ne permettant pas de définir une transition identique dans toutes les directions. La méthode présentée permet d'ajouter des classes lorsque les zones de transition sont trop étendues plutôt que de fixer une frontière arbitrairement
Actes de la conférence RFIA 2012
Lyon, France

Droits : info:eu-repo/semantics/OpenAccess
hal-00656502
https://hal.archives-ouvertes.fr/hal-00656502
https://hal.archives-ouvertes.fr/hal-00656502/document
https://hal.archives-ouvertes.fr/hal-00656502/file/rfia2012_submission_57.pdf

Éditeur(s) : HAL CCSD
Résumé : Session "Atelier IHMA"
National audience
Nous présentons une méthode pour la simulation des capacités d'évaluation de la dissemblance de stratégies de construction de dessins. Cette méthode exploite un descripteur structurel des procédures, l'OPSRL, ainsi qu'une mesure de dissemblance que nous présentons. La polyvalence des types de traitements qui peuvent être mis en oeuvre grâce à cette méthode est illustrée sur des jeux de tracés de formes géométriques qui ont été acquis en ligne au moyen de tablettes à digitaliser.
Actes de la conférence RFIA 2012
Lyon, France

Droits : info:eu-repo/semantics/OpenAccess
hal-00660983
https://hal.archives-ouvertes.fr/hal-00660983
https://hal.archives-ouvertes.fr/hal-00660983/document
https://hal.archives-ouvertes.fr/hal-00660983/file/IHMA_06_-_Ney_-_Remi.pdf

Éditeur(s) : HAL CCSD National Academy of Sciences
Résumé : International audience
All evolutionary biologists are familiar with evolutionary units that evolve by vertical descent in a tree-like fashion in single lineages. However, many other kinds of processes contribute to evolutionary diversity. In vertical descent, the genetic material of a particular evolutionary unit is propagated by replication inside its own lineage. In what we call introgressive descent, the genetic material of a particular evolutionary unit propagates into different host structures and is replicated within these host structures. Thus, introgressive descent generates a variety of evolutionary units and leaves recognizable patterns in resemblance networks. We characterize six kinds of evolutionary units, of which five involve mosaic lineages generated by introgressive descent. To facilitate detection of these units in resemblance networks, we introduce terminology based on two notions, P3s (subgraphs of three nodes: A, B, and C) and mosaic P3s, and suggest an apparatus for systematic detection of introgressive descent. Mosaic P3s correspond to a distinct type of evolutionary bond that is orthogonal to the bonds of kinship and genealogy usually examined by evolutionary biologists. We argue that recognition of these evolutionary bonds stimulates radical rethinking of key questions in evolutionary biology (e.g., the relations among evolutionary players in very early phases of evolutionary history, the origin and emergence of novelties, and the production of new lineages). This line of research will expand the study of biological complexity beyond the usual genealogical bonds, revealing additional sources of biodiversity. It provides an important step to a more realistic pluralist treatment of evolutionary complexity.
ISSN: 0027-8424

hal-01544803
https://hal.archives-ouvertes.fr/hal-01544803
DOI : 10.1073/pnas.1206541109

Éditeur(s) : HAL CCSD
Résumé : International audience
In this paper we describe the various steps of a method using local descriptors in order to classify freehand sketches, which are drawn in an unconstrained manner and recorded online. Our final objective is to simultaneously allow classification, comparison or clustering of those elements according to their Visuo spatial properties, their realization strategies, their structural or temporal parameters, or any combination of two or more of those parameters. The context and the motivation of this work-in-progress are described. First, we'll explain the detection step: we decided to use a mixed segmentation process. Indeed recognition methods based on local descriptor are not fully adapted to our types of shapes and to the one-line context. Later, we will show how we plan to give a global perception of the shape to our local descriptor by using segmentation results. Then we will introduce the local descriptor and how it is computing. Next we'll explain how we calculate the dissimilarity between two descriptors. Finally, we'll describe the approach we have chosen to determine the dissimilarity between shapes represented by sets of local descriptors.
Proceedings of the 14th Biennial Conference of the International Graphonomics Society

hal-00514930
https://hal.archives-ouvertes.fr/hal-00514930

Éditeur(s) : HAL CCSD
Résumé : International audience
We present the principle and the main steps of a new method for the visuo-spatial analysis of geometrical sketches recorded online. Visuo-spatial analysis is a necessary step for multi-level analysis. Multi-level analysis simultaneously allows classification, comparison or clustering of the constituent parts of a pattern according to their visuo-spatial properties, their procedural strategies, their structural or temporal parameters, or any combination of two or more of those parameters. The first results provided by this method concern the comparison of sketches to some perfect patterns of simple geometrical figures and the measure of dissimilarity between real sketches. The mean rates of good decision higher than 95% obtained are promising in both cases.
Proceedings of the 2010 International Conference on Image Processing, Computer Vision, & Pattern Recognition

Droits : info:eu-repo/semantics/OpenAccess
hal-00514929
https://hal.archives-ouvertes.fr/hal-00514929
https://hal.archives-ouvertes.fr/hal-00514929/document
https://hal.archives-ouvertes.fr/hal-00514929/file/IPCV2010_-_RRP.pdf
ARXIV : 1305.1520

Éditeur(s) : HAL CCSD
Résumé : International audience
Proceedings of the 15th International Conference of the Graphonomics Society - IGS 2011
Cancun, Mexico

hal-00604380
https://hal.archives-ouvertes.fr/hal-00604380

Éditeur(s) : HAL CCSD
Résumé : A standard approach for large scale image classification involves high dimensional features and Stochastic Gradient Descent algorithm (SGD) for the minimization of classical Hinge Loss in the primal space. Although complexity of Stochastic Gradient Descent is linear with the number of samples these method suffers from slow convergence. In order to cope with this issue, we propose here a Stochastic Low-Rank Newton Descent SLND for minimization of any calibrated loss in the primal space. SLND approximates the inverse Hessian by the best low-rank approximation according to squared Frobenius norm. We provide core optimization for fast convergence. Theoretically speaking, we show explicit convergence rates of the algorithm using these calibrated losses, which in addition provide working sets of parameters for experiments. Experiments are provided on the SUN, Caltech256 and ImageNet databases, with simple, uniform and efficient ways to tune remaining SLND parameters. On each of these databases, SLND challenges the accuracy of SGD with a speed of convergence faster by order of magnitude.
https://hal.archives-ouvertes.fr/hal-00825414

hal-00825414
https://hal.archives-ouvertes.fr/hal-00825414
https://hal.archives-ouvertes.fr/hal-00825414/document
https://hal.archives-ouvertes.fr/hal-00825414/file/TechnicalReport.pdf

Éditeur(s) : HAL CCSD Federation of American Society of Experimental Biology
Résumé : International audience
The innate immune system is able to detect bacterial LPS through the pattern recognition receptor CD14, which delivers LPS to various TLR signaling complexes that subsequently induce intracellular proinflammatory signaling cascades. In a previous study, we showed the overproduction of the soluble form of CD14 (sCD14) by macrophages from patients with cystic fibrosis (CF). CF is an autosomal recessive disorder that is caused by mutations in the gene that encodes the CFTR protein and is characterized by persistent inflammation. Macrophages play a significant role in the initial stages of this disease due to their inability to act as suppressor cells, leading to chronic inflammation in CF. In this work, we investigated the origin of sCD14 by human macrophages and studied the effect of sCD14 on the production of inflammatory cytokine/chemokine. Our data indicate that sCD14 stimulate proinflammatory cytokine/chemokine production in a manner that is independent of LPS but dependent on the TLR-4/CD14 membrane complex, NF-κB, and the inflammasome. Therefore, sCD14, overproduced by CF macrophages, originates primarily from the endocytosis/exocytosis process and should be considered to be a danger-associated molecular pattern. This elucidation of the origin and inflammation-induced mechanisms associated with sCD14 contributes to our understanding of maintained tissue inflammation.
ISSN: 0892-6638

hal-01526434
https://hal-univ-rennes1.archives-ouvertes.fr/hal-01526434
https://hal-univ-rennes1.archives-ouvertes.fr/hal-01526434/document
https://hal-univ-rennes1.archives-ouvertes.fr/hal-01526434/file/Soluble%20CD14%20acts%20as%20a%20DAMP%20in%20human-accepted.pdf
PUBMED : 28122919

Éditeur(s) : HAL CCSD
Résumé : International audience
“Affective Computing is computing that relates to, arises from, or deliberately influences emotions” (R.W. Picard, MIT Press, 1997). Specifically, psychosocial equilibrium is now considered as a major issue for both individual and public health. Predicting human emotions via non-intrusive methods is a great challenge triggered by the rise of intelligent assisting systems. In order to determine the emotional state of a subject having a potential psychosocial disorder, it is required to process both neutral and expressive emotions. A computer-engineered method aiming to provide emotional information from videos is proposed. An original approach for affect classification and its interpretation is presented.The database is part of the HUMAINE database, and consists of a large volume of video clips, each displaying individuals, depressed or not, talking with a conversational agent. Both the high-dimensionality of data and its temporal aspect require a pre-training that consists of cleaning, sorting and feature selection, in order to make relevant information usable for further processing by MachineLearning methods. Next step consists in classifying the affective state of subjects according to their positivity or negativity (Russell's model, 1980). By means of six independent classifiers, each running a Support Vector Machine algorithm, separate performances are obtained for each class of emotions. The results reveal an interesting “neutral area phenomenon”, where emotions tend to be harder to be detected.Though more frequent, neutral emotions seem to carry less discriminant information than expressive ones. This issue will be a key for designing a tuning step in further research.
Caribbean Academy Of Science
Deshaies, France

hal-01408029
https://hal.univ-antilles.fr/hal-01408029

Éditeur(s) : HAL CCSD Elsevier
Résumé : International audience
ISSN: 0262-8856

hal-01279721
https://hal.archives-ouvertes.fr/hal-01279721

Éditeur(s) : HAL CCSD Springer Verlag
Résumé : International audience
ISSN: 1433-7541

hal-01279720
https://hal.archives-ouvertes.fr/hal-01279720

Éditeur(s) : HAL CCSD
Résumé : Session "Posters"
National audience
Nous présentons une méthode permettant la mesure complète de la pose de la tête lors de captures vidéo monoculaires. Ce procédé utilise un modèle de visage et de texture précis. Le modèle facial, construit à partir d'une distribution de scans 3D haute résolution de visages, permet de générer des densités adaptées aux résolutions effectives. L'étape d'initialisation de la position du modèle est cruciale car elle conditionne le suivi ultérieur de la pose. Nous proposons ainsi d'utiliser une métrique composite construite sur des textures extraites de trois bases de données différentes. L'algorithme est alors capable de suivre le visage dans de vastes plages de poses avec une grande précision. Nous proposons des séquences vidéo de test dotées de vérité-terrain de pose précise et indépendante (précision de 0.1 degrés RMS)
Actes de la conférence RFIA 2012
Lyon, France

hal-00656565
https://hal.archives-ouvertes.fr/hal-00656565
https://hal.archives-ouvertes.fr/hal-00656565/document
https://hal.archives-ouvertes.fr/hal-00656565/file/rfia2012_submission_118.pdf