Éditeur(s) : HAL CCSD Hong Kong : World Academic Publishing
Résumé : International audience
In automatic image analysis, when using natural images, it is important to select the elements in a scene that can lead to correct characterization. Extraneous elements can result in poor values in the parameter evaluation for recognition. Irrelevant elements also increase the processing time, because they can be used incorrectly by the algorithms for characterization or identification. The goal of this paper is to offer identification algorithms and characterizations that will allow effective recognition. We base our procedure on the photographer focus areas to identify those relevant elements in a scene. In this work, we propose a procedure to select the relevant shape from natural images. The treatment uses focal depth with edge detectors. The resulting points are combined with a region partition to obtain relevant shapes.
ISSN: 2309-0650

hal-01278550
https://hal.archives-ouvertes.fr/hal-01278550

Éditeur(s) : HAL CCSD IEEE
Résumé : International audience
Recent works display that large scale image classification problems rule out computationally demanding methods. On such problems, simple approaches like k-NN are affordable contenders, with still room space for statistical improvements under the algorithmic constraints. A recent work showed how to leverage k-NN to yield a formal boosting algorithm. This method, however, has numerical issues that make it not suited for large scale problems. We propose here an Adaptive Newton-Raphson scheme to leverage k-NN, N³, which does not suffer these issues. We show that it is a boosting algorithm, with several key algorithmic and statistical properties. In particular, it may be sufficient to boost a subsample to reach desired bounds for the loss at hand in the boosting framework. Experiments are provided on the SUN, and Caltech databases. They confirm that boosting a subsample -- sometimes containing few examples only -- is sufficient to reach the convergence regime of N³. Under such conditions, N³ challenges the accuracy of contenders with lower computational cost and lower memory requirement.
IEEE International Workshop on Machine Learning for Signal Processing
Southampton, United Kingdom

hal-00959125
https://hal.archives-ouvertes.fr/hal-00959125
https://hal.archives-ouvertes.fr/hal-00959125/document
https://hal.archives-ouvertes.fr/hal-00959125/file/BNNB_mlsp2013.pdf

Éditeur(s) : Université de Bretagne Occidentale
Résumé : Study of mineral suspended particulate matter (SPM) is essential to understand ecosystem dynamics. SPM influences light penetration which is closely related to primary production and carries pollutants and nutrients within the water masses. This work aims to study near-surface mineral SPM dynamics in the English Channel using two complementary methods: remote-sensing by satellites and numerical modelling. Firstly, MODIS and MERIS satellite images processed by the IFREMER semi-analytical algorithm and describing non-algal SPM, are analysed to build three statistical models which are able to estimate near-surface SPM using simple variables, e.g., tidal coefficient, significant wave height and mean chlorophyll-a concentration. Secondly, the three-dimensional hydrosedimentary model ROMS (Regional Ocean Modeling System) is applied in the English Channel. It considers tidal forcing along open boundaries, wind stress at the sea-surface, wave-current interactions in the bottom boundary layer and realistic heterogeneous bottom sediments. Results of the more sophisticated statistical model is also used to impose SPM concentrations at open boundaries. Sensitivity tests on SPM boundary conditions show the importance of prescribing realistic sediment concentrations to improve the model predictions. Biological influences on mineral SPM are also tested using settling velocity and critical shear stress for erosion which vary with seasons. Both statistical and numerical models highlight and locate respective influences of tide, waves and biology on mineral SPM. Numerical predictions indicate that near-surface SPM is mainly composed of silts (>70%). Near-surface suspended silts come largely from seabed resuspension alongshore in the Eastern Channel and from the open boundaries of our limited-area coastal model at the West of Cotentin Peninsula and offshore. Near-surface SPM variations are investigated at semi-diurnal, neap-spring tidal cycle time scales and during a wave event. Finally, dynamics of sediments in the turbidity maximum zone around the Isle of Wight is examined depending on hydrodynamical conditions in order to distinguish the role played by advection and local erosion/settling processes in this area. Keywords: Suspended particulate matter, satellite, remote-sensing, modelling, English Channel, sediments, turbidity, tide, wave
L'étude des matières en suspension (MES) minérales est essentielle pour comprendre le fonctionnement des écosystèmes en Manche : les MES influencent la pénétration de la lumière dans la colonne d'eau, paramètre clé pour la production biologique, et sont susceptibles de transporter des polluants et nutriments. Lors de ce travail, la dynamique des MES minérales de surface en Manche est étudiée à l'aide de deux méthodes complémentaires : l'observation par satellite et la modélisation numérique tridimensionnelle. Dans un premier temps, les images satellite MODIS et MERIS, traitées par l'algorithme semi-analytique développé par l'IFREMER et représentant les MES non-algales de surface, sont analysées afin d'élaborer trois modèles statistiques permettant d'estimer la MES de surface en fonction de variables basiques telles que le coefficient de marée, la hauteur significative des vagues et la concentration moyenne en chlorophylle-a. Dans un deuxième temps, le modèle hydrosédimentaire multiclasse tridimensionnel ROMS (Regional Ocean Modeling System) est implémenté en Manche. Il prend en considération le forçage par la marée aux frontières ouvertes, la contrainte exercée par le vent à la surface de la mer, l'interaction houle-courant en couche limite de fond et l’hétérogénéité spatiale réaliste du lit sédimentaire. Les résultats du modèle statistique le plus sophistiqué sont également utilisés pour forcer la concentration en MES aux frontières ouvertes du domaine de calcul. Des tests de sensibilité sur les conditions limites montrent l'importance de considérer de manière réaliste les sédiments entrant en Manche par les frontières ouvertes lors de la modélisation numérique. L'influence de la biologie sur les MES minérales est testée en utilisant une formulation de la vitesse de chute et de la contrainte critique d'érosion variant saisonnièrement. Grâce à ces deux outils associés, les influences respectives de la marée, de la houle et de la biologie sur les MES minérales sont mises en avant et localisées en Manche. Le modèle numérique prédit que les MES de surface sont composées principalement de silts (>70%). Les silts en suspension en surface proviennent majoritairement du lit sédimentaire près des côtes en Manche orientale et des frontières ouvertes du domaine de calcul à l'ouest de la presqu'île du Cotentin et au large en Manche centrale et orientale. Les variations des MES de surface sont ensuite étudiées aux échelles des cycles de marée semi-diurnes, vive-eau/morte-eau et autour d'un épisode de houles. Enfin, l'évolution de la concentration de sédiments dans la zone de forte turbidité autour de l'île de Wight peut être analysée suivant les conditions hydrodynamiques afin d'évaluer les rôles respectifs des processus locaux d'érosion/dépot et d'advection dans la génération de ces turbidités remarquables. Mots-clés : Matière en suspension, télédétection, satellite, modélisation, Manche, sédiments, turbidité, marée, houle

Droits : 2013 UBO
http://archimer.ifremer.fr/doc/00157/26783/24884.pdf
http://archimer.ifremer.fr/doc/00157/26783/

Éditeur(s) : HAL CCSD
Résumé : International audience
In this paper, we investigate image segmentation by region merging. Given any similarity measure between regions , satisfying some weak constraints, we give a general predicate for answering if two regions are to be merged or not during the segmentation process. Our predicate is generic and has six properties. The first one is its inde-pendance with respect to the similarity measure, that leads to a user-independant and adaptative predicate. Second, it is non-parametric, and does not rely on any assumption concerning the image. Third, due to its weak constraints, knowledge may be included in the predicate to fit better to the user's behaviour. Fourth, provided the similarity is well-chosen by the user, we are able to upperbound one type of error made during the image segmentation. Fifth, it does not rely on a particular segmentation algorithm and can be used with almost all region-merging algorithms in various application domains. Sixth, it is calculated quickly, and can lead with appropriated algorithms to very efficient segmentation.
10th IEEE International Conference on Tools with Artificial Intelligence
Taipei, Taiwan

lirmm-01168371
https://hal-lirmm.ccsd.cnrs.fr/lirmm-01168371
https://hal-lirmm.ccsd.cnrs.fr/lirmm-01168371/document
https://hal-lirmm.ccsd.cnrs.fr/lirmm-01168371/file/FioNocICTAI98_Image%20Segmentation%20Using.pdf
DOI : 10.1109/TAI.1998.744885

Éditeur(s) : HAL CCSD
Résumé : Ce travail traite de " l'identification de plantes à partir d'images numériques, par champs de gradients ", provenant de capteurs de types variés. Nous avons choisi des méthodes de type global, afin de pré-segmenter efficacement l'image d'une plante. Parmi ces méthodes, celle du Mean Shift permet de s'acquitter des tâches de paramétrisation, mais présente l'inconvénient d'une forte complexité de calcul. Nous proposons deux méthodes pour résoudre ce problème : le Mean Shift avec changement d'échelle et le Mean Shift Markovien dont le but est de limiter le traitement systématique de tous les pixels par la partie Mean Shift. L'application de ces méthodes permet d'obtenir une partition de l'image dans laquelle on recherche les groupes de pixels, matérialisant les parties d'une plante focalisée par le photographe. Pour mettre en évidence ces zones, nous exploitons : la netteté des formes, la position et la densité spatiale des objets dans une image, afin de construire des masques qui représentent les parties d'une plante. À partir de ces masques, nous extrayons un ensemble de caractéristiques issu de la botanique descriptive, regroupé en trois catégories : la colorimétrie, la morphologie et le structurel. Nos choix méthodologiques sont expérimentés sur une base de données d'images, pour obtenir un ensemble de valeurs de caractéristiques appelées descripteurs. Ces descripteurs sont fournis à un système de classeurs basé sur des algorithmes génétiques, afin d'obtenir des règles désignant chaque type de plantes. L'objectif de ces règles est d'associer un score de ressemblance à chaque image de la base de données, en fonction de l'image soumise comme requête au système de reconnaissance.
Journées d'Études et de recherche des Antilles et de la Guyane
Pointe-à-Pitre, Guadeloupe

hal-00637788
https://hal.archives-ouvertes.fr/hal-00637788

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 Oxford University Press (OUP)
Résumé : International audience
Several studies, both theoretical and experimental, show that sedimentary rocks have a fractal pore-grain interface. In this paper a computer simulated 3-D sedimentary rock structure generated by the Relaxed Ballistic Bidisperse Deposition Model (RBBDM), is investigated to characterize the micro structure of its pores. The pore volume and the rock-pore interface show the same fractal dimension indicating that the pore volume is a fractal. The two point density correlation is computed for the pore space and the results compare favourably with the range reported from experiments. An array of 2-D X-ray tomography micrograph sections of a real sedimentary rock, an oolitic limestone (pure calcite) from the Mondeville formation of Middle Jurassic age (Paris Basin, France), was used to generate a 3-D bitmap. The 3-D real rock sample generated in this manner, was analysed for similar studies as the simulated structure. The results were compared with those obtained from simulation. The simulation results agree qualitatively with the real rock sample. Diffusion through the connected pore space of the simulated structure was studied using a random walk algorithm and the results compared with the similar simulation study done on the 3-D oolitic limestone specimen. In both cases diffusion was found to be anomalous indicating that the sedimentary rock has a fractal geometry. The favourable comparability of results between the simulated and real rock supports the usefulness of the model of sedimentary rock generation which can be applicable to transport phenomena.
ISSN: 0956-540X

hal-00823061
https://hal.archives-ouvertes.fr/hal-00823061
DOI : 10.1093/gji/ggs084

Éditeur(s) : HAL CCSD Kluwer
Résumé : This paper presents some considerations about hybrid optimization algorithms useful to optimize real complex system. The given indication could be used by readers to conceive hybrid algorithms. These considerations have been deducted from a concrete application case: the satellites constellations design problem. But each of the advanced techniques proposed in this paper are considered in a more general way to solve other problems. This problem is used to illustrate the techniques along the paper because it is grouping many characteristics (di culties) of contemporary real complex systems: the size and the characteristics of the search space engendered by a combinatorial problem; The irregularity of the criterions; The mathematical and physical heterogeneity of parameters forbids the use of classical algorithms; The evaluation of a solution, which uses a time consuming simulation; A need of accurate values. More details are available in previous papers ([?], [?], [?], [?]). For these reasons, we could learn a lot from this experiment in order to detach hybrid techniques usable for problems having close characteristics. This paper presents the historic leading to the current algorithm, the modeling of the complex system and the sophisticated algorithm proposed to optimize it. Application cases and ways to built signi cant tests of hybrid algorithm are also given.
International Federation for Information Processing

hal-00602279
https://hal.archives-ouvertes.fr/hal-00602279

Éditeur(s) : HAL CCSD
Résumé : This paper presents the experimental results obtained for an automatic detection of ships in High Resolution optical satellite images. Our images are panchromatic SPOT 5 images, whose resolution is 5m per pixel. Our detection method is part of an industrial project of maritime surveillance. It is based on the Bayesian decision theory and does not need any preprocessing.The detection is composed of three stages. The first one is a pre-detection of targets that gives us candidates. The second one is a precise segmentation of each candidate. The third one is a classification of candidates into three classes : real small targets, real big targets and false alarms. The two first stages are based on the bayesian theory, using a very simple image model that leads to very fast algorithms. For now, the classification is based on multivariate decision tree. Finally, the overall results of the method are given for a set of images, as close as possible to the operational conditions and show the performances of the proposed method (detection of almost all the ships, false alarms due to crests of waves and an acceptable computing time).
Journée Nationale de Photogrammétrie et Télédétection

hal-00601854
https://hal.archives-ouvertes.fr/hal-00601854

Éditeur(s) : HAL CCSD IEEE
Résumé : International audience
This paper proposes a novel automated approach for the categorization of cells in fluorescence microscopy images. Our supervised classification method aims at recognizing patterns of unlabeled cells based on an annotated dataset. First, the cell images need to be indexed by encoding them in a feature space. For this purpose, we propose tailored bio-inspired features relying on the distribution of contrast information. Then, a supervised learning algorithm is proposed for classifying the cells. We carried out experiments on cellular images related to the diagnosis of autoimmune diseases, testing our classification method on the HEp-2 Cells dataset of Foggia et al (CBMS 2010). Results show classification precision larger than 96% on average, thus confirming promising application of our approach to the challenging application of cellular image classification for computer-aided diagnosis.
ICPR 2012
Tsukuba, Japan

hal-00958856
https://hal.archives-ouvertes.fr/hal-00958856
https://hal.archives-ouvertes.fr/hal-00958856/document
https://hal.archives-ouvertes.fr/hal-00958856/file/icpr_2012.pdf

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 : International audience
This paper presents some considerations about hybrid optimization algorithms useful to optimize real complex system. The given indication could be used by readers to conceive hybrid algorithms. These considerations have been deducted from a concrete application case: the satellites constellations design problem. But each of the advanced techniques proposed in this paper are considered in a more general way to solve other problems. This problem is used to illustrate the techniques along the paper because it is grouping many characteristics (di culties) of contemporary real complex systems: the size and the characteristics of the search space engendered by a combinatorial problem; The irregularity of the criterions; The mathematical and physical heterogeneity of parameters forbids the use of classical algorithms; The evaluation of a solution, which uses a time consuming simulation; A need of accurate values. More details are available in previous papers ([?], [?], [?], [?]). For these reasons, we could learn a lot from this experiment in order to detach hybrid techniques usable for problems having close characteristics. This paper presents the historic leading to the current algorithm, the modeling of the complex system and the sophisticated algorithm proposed to optimize it. Application cases and ways to built signi cant tests of hybrid algorithm are also given.
HPCSE Proceedings
Toulouse, France

hal-00602264
https://hal.archives-ouvertes.fr/hal-00602264

Éditeur(s) : HAL CCSD
Résumé : International audience
The k-nearest neighbors (k-NN) classification rule is still an essential tool for computer vision applications, such as scene recognition. However, k-NN still features some major drawbacks, which mainly reside in the uniform voting among the nearest prototypes in the feature space. In this paper, we propose a new method that is able to learn the "relevance" of prototypes, thus classifying test data using a weighted k-NN rule. In particular, our algorithm, called Multi-class Leveraged k-nearest neighbor (MLNN), learns the prototype weights in a boosting framework, by minimizing a surrogate exponential risk over training data. We propose two main contributions for improving computational speed and accuracy. On the one hand, we implement learning in an inherently multiclass way, thus providing significant computation time reduction over one-versus-all approaches. Furthermore, the leveraging weights enable effective data selection, thus reducing the cost of k-NN search at classification time. On the other hand, we propose a kernel generalization of our approach to take into account real-valued similarities between data in the feature space, thus enabling more accurate estimation of the local class density. We tested MLNN on three datasets of natural images. Results show that MLNN significantly outperforms classic k-NN and weighted k-NN voting. Furthermore, using an adaptive Gaussian kernel provides significant performance improvement. Finally, the best results are obtained when using MLNN with an appropriate learned metric distance.
Proceedings of the 10th Asian Conference on Computer Vision, ACCV 2010, November 8-12, 2010, Queenstown, New Zealand
Queenstown, New Zealand

hal-00664606
https://hal.inria.fr/hal-00664606
https://hal.inria.fr/hal-00664606/document
https://hal.inria.fr/hal-00664606/file/mlnn_accvfinal.pdf

Éditeur(s) : HAL CCSD Institute of Electrical and Electronics Engineers
Résumé : International audience
Tailoring nearest neighbors algorithms to boosting is an important problem. Recent papers study an approach, UNN, which provably minimizes particular convex surrogates under weak assumptions. However, numerical issues make it necessary to experimentally tweak parts of the UNN algorithm, at the possible expense of the algorithm's convergence and performance. In this paper, we propose a lightweight alternative algorithm optimizing proper scoring rules from a very broad set, and establish formal convergence rates under the boosting framework that surprisingly compete with those known for UNN. It is an adaptive Newton-Raphson algorithm, which belongs to the same lineage as the popular Gentle Adaboost. To the best of our knowledge, no such boosting-compliant convergence rates were previously known for these algorithms. We provide experiments on a dozen domains, including the challenging Caltech and SUN computer vision databases. They display that GNNB significantly outperforms UNN, both in terms of convergence rate and quality of the solution obtained, and GNNB provides a simple and efficient contender to techniques that can be used on very large domains, like stochastic gradient descent -- for which little is known to date. Experiments include a divide-and-conquer improvement of GNNB which exploits the link with proper scoring rules optimization.
ISSN: 0162-8828

hal-00958809
https://hal.archives-ouvertes.fr/hal-00958809
https://hal.archives-ouvertes.fr/hal-00958809/document
https://hal.archives-ouvertes.fr/hal-00958809/file/PAMI.pdf
DOI : 10.1109/TPAMI.2014.2307877

Éditeur(s) : HAL CCSD
Résumé : International audience
In this paper we propose a hybrid approach using mutual information and multi-objective optimization for feature subset selection problem. The hybrid aspect is due to the sequence of a filter method and a wrapper method in order to take advantages of both. The filter method reduces the exploration space by keeping subsets having good internal properties and the wrapper method chooses among the remaining subsets with a classification performances criterion. In the filter step, the subsets are evaluated in a multi-objective way to ensure diversity within the subsets. The evaluation is based on the mutual information to estimate the dependency between features and classes and the redundancy between features within the same subset. We kept the non-dominated (Pareto optimal) subsets for the second step. In the wrapper step, the selection is made according to the stability of the subsets regarding classification performances during learning stage on a set of classifiers to avoid the specialization of the selected subsets for a given classifiers. The proposed hybrid approach is experimented on a variety of reference data sets and compared to the classical feature selection methods FSDD and mRMR. The resulting algorithm outperforms these algorithms and the computation complexity remains acceptable even if it increases with regards to these two fast selection methods.
Journal of Informatics and Mathematical Sciences

hal-01375982
https://hal.archives-ouvertes.fr/hal-01375982

Éditeur(s) : HAL CCSD
Résumé : The research presented in this thesis aims at determining the possibility of partial or total automatic ship detection in High Resolution optical satellite images. At first, we expose the industrial context of our study, which fixes the constraints in terms of performances and processing time. Then, we present the available satellite data for our study (SPOT 5 panchromatic images of 5m per pixel resolution). Then we detail our approach for ship detection : the first stage is a pre-detection of targets that gives us candidates. The second stage is a precise segmentation of each candidate and the third stage is a classification of candidates. Our pre-detection method is based on the Bayesian decision theory. This pre-detection supplies a set of regions of interests (ROI). Among them, we have ships, but also false alarms due to clouds and to crests of waves. The purpose of the following stage is to eliminate the false alarms and to characterize the pre-detected ships. Then, we detail the method used to segment candidates. It uses active contours and also uses the Bayesian decision theory. After a brief description of the characteristics extracted for the classication of ships, we present the classifier that we used. We complete the ship characterization by presenting two methods of wave ship detection. The first approach uses active contours and the second is based on an a contrario approach. We compare the results of these two approaches. To complete the study and replace it in the industrial and operational context, we present experiments of the complete algorithm on complete images to estimate the performances and the processing time of the whole method.
Les travaux présentés dans cette thèse ont pour but de déterminer la possibilité d'automatiser, partiellement ou totalement, la détection de bateaux à partir d'images satellitaires optiques haute résolution. Nous exposons tout d'abord le contexte industriel dans lequelle s'insère notre étude, qui fixe les contraintes en termes de performances et de temps de traitement. Puis, nous présentons les données satellitaires disponibles dans le cadre de notre étude (images SPOT 5 panchromatique de résolution 5m). Après avoir illustré le contenu des scènes, nous présentons la démarche adoptée pour la détection des navires : une phase de pré-détection suivie d'une étape de segmentation et d'une étape de classification. La méthode de pré-détection utilisée s'appuie sur la théorie bayésienne de la décision. Après avoir présenté le modèle classique, nous discutons du modèle que nous avons utilisé. Cette pré-détection fournit un ensemble de régions d'intérêts. Parmi elles, nous n'avons pas uniquement des bateaux, nous avons également des fausses alarmes dues aux nuages et aux crêtes de vagues. Le but de l'étape suivante est à la fois d'éliminer les fausses alarmes et de caractériser les bateaux pré-détectés. Ensuite, nous détaillons la méthode utilisée pour segmenter les bateaux. Elle s'appuie sur des contours actifs et utilise également la théorie bayésienne de la décision. Après avoir décrit les caractéristiques extraites pour la classification des bateaux, nous présentons le classifieur utilisé. Nous complétons la caractérisation des navires en présentant deux méthodes de détection des sillages différentes des méthodes classiques. La première approche s'appuie sur l'utilisation de contours actifs et la seconde est basée sur une approche a contrario. Nous comparons les résultats des ces deux approches. An de compléter l'étude et se replacer dans le contexte industriel et opérationnel de la thèse, nous présentons des expérimentations de l'ensemble de l'algorithme sur des images complètes pour évaluer les performances et le temps de traitement.
https://tel.archives-ouvertes.fr/tel-00560119

tel-00560119
https://tel.archives-ouvertes.fr/tel-00560119
https://tel.archives-ouvertes.fr/tel-00560119/document
https://tel.archives-ouvertes.fr/tel-00560119/file/These_NadiaProia_Version_Finale.pdf

Éditeur(s) : HAL CCSD
Résumé : Joint inversions are now commonly used in Earth Sciences. They have been developed to better understand Earth structure since they provide more constraints on the inverted parameters. We propose a new process to simultaneously invert several data sets in order to better image 3-D crustal and upper mantle structures. Our inversion uses 3 kind of data which present a good complementarity : (1) P receiver functions to provide Moho depth variations, (2) teleseismic delay times of P waves to retrieve velocity anomalies in the crust and upper mantle, (3) gravity anomalies to image density variations at lithospheric scale. In our scheme, receiver functions are first inverted. The resulting Moho depths are then interpolated and incorporated as a priori information into the joint inversion of teleseismic delay times and gravity anomalies process. In our new approach, we perform a model space search for Moho variations, P-velocity and density structure to find acceptable fit to the three data sets. In order to preferentially sample the good data fit region, we chose the Neighborhood algorithm of Sambridge to optimistically survey the model space. We model the delay times with a 3-D raytracing using velocity nodes evenly spaced and linked with density nodes via a linear relationship. This relationship can evolve through depth, simulating its pressure dependance. We test our scheme on synthetic examples, and apply the method in a continental rifting region where deformation processes are complex and badly known. This joint inversion thus provides more constraints on the inverted parameters, and allows us to better image crust - mantle interactions in the case of the Hangai dome (Central Mongolia). Indeed, in this area, we expect an asthenospheric upwelling to dynamically support the topography, but we image a density - velocity anomaly in the lower crust. The effects on topography and the origin (composition vs. thermal) of this anomaly are still under debate.
Les inversions conjointes sont maintenant communément utilisées en Sciences de la Terre. Elles ont été développées afin d'améliorer notre connaissance et compréhension de la structure interne de la Terre puisqu'elles apportent de plus en plus de contraintes sur les paramètres inversés. Dans cette thèse, nous proposons un nouveau processus d'inversion conjointe qui prend en compte trois paramètres différents et qui mène à l'obtention d'un modèle lithosphérique tridimensionnel. Notre méthode utilise trois types de données gravimétriques et sismologiques qui présentent une bonne complémentarité : (1) Les fonctions récepteur P afin d'obtenir les variations de profondeur du Moho, (2) les délais de temps d'arrivées P des téléséismes pour retrouver les anomalies de vitesse dans la croûte et le manteau supérieur, (3) les anomalies gravimétriques qui donnent accès aux variations de densité à l'échelle lithosphérique. Dans le schéma d'inversion proposé, nous inversons d'abord les fonctions récepteur. Les variations de profondeur du Moho qui en résultent sont alors interpolées puis incorporées comme information a priori dans le processus d'inversion conjointe gravimétrie - tomographie télésismique. Que ce soit pour les fonctions récepteur ou pour la partie conjointe, notre méthode d'inversion se base sur un algorithme de recherche afin de trouver la structure qui minimise l'écart aux trois types de données. Nous avons opté pour l'algorithme de voisinage qui présente l'avantage de concentrer sa recherche dans les régions qui minimisent l'écart aux données sans toutefois abandonner les autres régions afin de limiter les risques de convergence vers un minimum local. Nous utilisons un modèle 3-D constitué de noeuds de vitesse afin de modéliser les délais de temps. Ces noeuds de vitesse sont liés aux noeuds de densité grâce à une relation linéaire entre vitesse et densité. Cette relation peut évoluer avec la profondeur de manière à simuler les effets dus aux variations de pression. Nous avons réalisé plusieurs tests synthétiques afin d'évaluer le comportement de notre nouvelle méthode. Ils ont permis de montrer que les variations du Moho ainsi que les anomalies de vitesse et densité sont bien retrouvées. Ils ont également montré l'importance de la prise en compte de ces trois types de données dans un même schéma d'inversion, notamment en ce qui concerne la faculté de discerner les variations d'interfaces des anomalies volumiques. Nous avons ensuite appliqué la méthode dans une région continentale en phase de rifting où les processus de déformation sont complexes et mal connus. Notre nouvelle méthode a apporté davantage de contraintes sur les paramètres inversés et a permis d'obtenir un modèle correctement résolu de la croûte et du manteau supérieur sous le dôme de Hangai (Mongolie Centrale). Une anomalie asthénosphérique sous le dôme est souvent invoquée pour expliquer sa topographie élevée mais c'est une anomalie (en vitesse et densité) crustale (croûte inférieure) qui s'exprime dans nos modèles. Ses effets sur la topographie et son origine (compositionnelle et/ou thermique) sont encore débattus et nécessiteraient des études plus poussées (paramétriques, VP =VS, ...).
https://tel.archives-ouvertes.fr/tel-00783524

tel-00783524
https://tel.archives-ouvertes.fr/tel-00783524
https://tel.archives-ouvertes.fr/tel-00783524/document
https://tel.archives-ouvertes.fr/tel-00783524/file/Basuyau2011.pdf

Éditeur(s) : HAL CCSD Elsevier
Résumé : International audience
Voting rules relying on k-nearest neighbors (k-NN) are an effective tool in countless many machine learning techniques. Thanks to its simplicity, k-NN classification is very attractive to practitioners, as it enables very good performances in several practical applications. However, it suffers from various drawbacks, like sensitivity to "noisy" instances and poor generalization properties when dealing with sparse high-dimensional data. In this paper, we tackle the k-NN classification problem at its core by providing a novel k-NN boosting approach. Namely, we propose a supervised learning algorithm, called Universal Nearest Neighbors (UNN), that induces a leveraged k-NN rule by globally minimizing a surrogate risk upper bounding the empirical misclassification rate over training data. Interestingly, this surrogate risk can be arbitrary chosen from a class of Bregman loss functions, including the familiar exponential, logistic and squared losses. Furthermore, we show that UNN allows to efficiently filter a dataset of instances by keeping only a small fraction of data. Experimental results on the synthetic Ripley's dataset show that such a filtering strategy is able to reject "noisy" examples, and yields a classification error close to the optimal Bayes error. Experiments on standard UCI datasets show significant improvements over the current state of the art.
ISSN: 0925-2312

hal-00664462
https://hal.inria.fr/hal-00664462

Éditeur(s) : HAL CCSD
Résumé : International audience
In this paper, we use a multifractal approach based on the computation of two spectrums for image analysis and texture segmentation problems. The two spectrums are the Legendre Spectrum, determined by classical methods, and the Large Deviation Spectrum, determined by kernel density estimation. We propose a way for the fusion of these two spectrums to improve textured image segmentation results. An unsupervised k-means is used as clustering approach for the texture classification. The algorithm is applied on mosaic image built using IKONOS images and various natural textures from the Brodatz album. The segmentation obtained with our approach gives better results than the application of each spectrum separately.
CIARP Proceedings
Cancun, Mexico

hal-00602268
https://hal.archives-ouvertes.fr/hal-00602268

Éditeur(s) : HAL CCSD
Résumé : International audience
This paper deals with hybrid optimization take different methods to solve a complex problem. As if the choice of concerning methods is important, the way to combine them is determinant for an ef cient search. The article deals with cooperation strategy between the different layers of an hybrid optimization algorithm. We describ the way it is articulated for an Satellite Constellation Optimization problem. This cooperation allows a dynamical adaptation of the optimization algorithm parameters according to the mutual advance of each search
Journal Européen des Systèmes Avancés

hal-00602267
https://hal.archives-ouvertes.fr/hal-00602267