Publications about Classification

Result of the query in the list of publications :

• 3 Articles
• 7 PhD Thesis and Habilitations
• 12 Conference articles
• 12 Technical and Research Reports

new search | get the bibtex

navigation : 1 - 2 display : 10 per page | 20 per page | 50 per page

3 Articles

1 - Unsupervised amplitude and texture classification of SAR images with multinomial latent model. K. Kayabol and J. Zerubia. IEEE Trans. on Image Processing, 22(2): pages 561-572, February 2013. Keywords : COSMOSkyMed, Classification EM, High resolution SAR, Jensen-Shannon criterion, Classification, Multinomial logistic. @ARTICLE{KorayTIP2013, author = {Kayabol, K. and Zerubia, J.}, title = {Unsupervised amplitude and texture classification of SAR images with multinomial latent model}, year = {2013}, month = {February}, journal = {IEEE Trans. on Image Processing}, volume = {22}, number = {2}, pages = {561-572}, url = {http://hal.inria.fr/hal-00745387}, keyword = {COSMOSkyMed, Classification EM, High resolution SAR, Jensen-Shannon criterion, Classification, Multinomial logistic} }

2 - A study of Gaussian mixture models of colour and texture features for image classification and segmentation. H. Permuter and J.M. Francos and I. H. Jermyn. Pattern Recognition, 39(4): pages 695--706, April 2006. Keywords : Classification, Segmentation, Texture, Colour, Gaussian mixture, Decison fusion. @ARTICLE{permuter_pr06, author = {Permuter, H. and Francos, J.M. and Jermyn, I. H.}, title = {A study of Gaussian mixture models of colour and texture features for image classification and segmentation}, year = {2006}, month = {April}, journal = {Pattern Recognition}, volume = {39}, number = {4}, pages = {695--706}, url = {http://dx.doi.org/10.1016/j.patcog.2005.10.028}, pdf = {ftp://ftp-sop.inria.fr/ariana/Articles/2006_permuter_pr06.pdf}, keyword = {Classification, Segmentation, Texture, Colour, Gaussian mixture, Decison fusion} } Abstract : The aims of this paper are two-fold: to define Gaussian mixture models of coloured texture on several feature paces and to compare the performance of these models in various classification tasks, both with each other and with other models popular in the literature. We construct Gaussian mixtures models over a variety of different colour and texture feature spaces, with a view to the retrieval of textured colour images from databases. We compare supervised classification results for different choices of colour and texture features using the Vistex database, and explore the best set of features and the best GMM configuration for this task. In addition we introduce several methods for combining the 'colour' and 'structure' information in order to improve the classification performance. We then apply the resulting models to the classification of texture databases and to the classification of man-made and natural areas in aerial images. We compare the GMM model with other models in the literature, and show an overall improvement in performance.

3 - Supervised Segmentation of Remote Sensing Images Based on a Tree-Structure MRF Model. G. Poggi and G. Scarpa and J. Zerubia. IEEE Trans. Geoscience and Remote Sensing, 43(8): pages 1901-1911, August 2005. Keywords : Classification, Segmentation, Markov Fields. @ARTICLE{ieeetgrs_05, author = {Poggi, G. and Scarpa, G. and Zerubia, J.}, title = {Supervised Segmentation of Remote Sensing Images Based on a Tree-Structure MRF Model}, year = {2005}, month = {August}, journal = {IEEE Trans. Geoscience and Remote Sensing}, volume = {43}, number = {8}, pages = {1901-1911}, pdf = {http://ieeexplore.ieee.org/iel5/36/32001/01487647.pdf?tp=&arnumber=1487647&isnumber=32001}, keyword = {Classification, Segmentation, Markov Fields} }

top of the page

7 PhD Thesis and Habilitations

1 - Shape recognition for image scene analysis. M. S. Kulikova. PhD Thesis, Universite de Nice - Sophia-Antipolis, December 2009. Keywords : tree crown , Classification, Shape, multiple object extraction, Marked point process, Shape prior. @PHDTHESIS{mkulikova_phd09, author = {Kulikova, M. S.}, title = {Shape recognition for image scene analysis}, year = {2009}, month = {December}, school = {Universite de Nice - Sophia-Antipolis}, url = {http://tel.archives-ouvertes.fr/docs/00/48/20/19/PDF/phd_mkulikova_2009.pdf}, keyword = {tree crown , Classification, Shape, multiple object extraction, Marked point process, Shape prior} } Résumé : Cette thèse est composée de deux parties principales. La première partie est dédiée au problème de la classification d’espèces d’arbres en utilisant des descripteurs de forme, en combainison ou non, avec ceux de radiométrie ou de texture. Nous montrons notamment que l’information sur la forme améliore la performance d’un classifieur. Pour ce faire, dans un premier temps, une étude des formes de couronnes d’arbres extraites à partir d’images aériennes, en infrarouge couleur, est eectuée en utilisant une méthodologie d’analyse de formes des courbes continues fermées dans un espace de formes, en utilisant la notion de chemin géodésique sous deux métriques dans des espaces appropriés : une métrique non-élastique en utilisant la reprèsentation par la fonction d’angle de la courbe, ainsi qu’une métrique élastique induite par une représentation par la racinecarée appelée q-fonction. Une étape préliminaire nécessaire à la classification est l’extraction des couronnes d’arbre. Dans une seconde partie, nous abordons donc le problème de l’extraction d’objets de forme complexe arbitraire, à partir d’images de télédétection à très haute résolution. Nous construisons un modèle fondé sur les processus ponctuels marqués. Son originalité tient dans sa prise en compte d’objets de forme arbitraire par rapport aux objets de forme paramétrique, e.g. ellipses ou rectangles. Les formes sélectionnées sont obtenues par la minimisation locale d’une énergie de type contours actifs avec diérents a priori sur la forme incorporé. Les objets de la configuration finale (optimale) sont ensuite sélectionnés parmi les candidats par une dynamique de naissances et morts multiples, couplée à un schéma de recuit simulé. L’approche est validée sur des images de zones forestières à très haute résolution fournies par l’Université d’Agriculture de Suède. Abstract : This thesis includes two main parts. In the first part we address the problem of tree crown classification into species using shape features, without, or in combination with, those of radiometry and texture, to demonstrate that shape information improves classification performance. For this purpose, we first study the shapes of tree crowns extracted from very high resolution colour aerial infra-red images. For our study, we choose a methodology based on the shape analysis of closed continuous curves on shape spaces using geodesic paths under the bending metric with the angle-function curve representation, and the elastic metric with the square root q-function representation. A necessary preliminary step to classification is extraction of the tree crowns. In the second part, we address thus the problem of extraction of multiple objects with complex, arbitrary shape from remote sensing images of very high resolution. We develop a model based on marked point processes. Its originality lies in its use of arbitrarily-shaped objects as opposed to parametric shape objects, e.g. ellipses or rectangles. The shapes considered are obtained by local minimisation of an active contour energy with weak and then strong shape prior knowledge included. The objects in the final (optimal) configuration are then selected from amongst these candidates by a multiple birth-and-death dynamics embedded in an annealing scheme. The approach is validated on very high resolution images of forest provided by the Swedish University of Agriculture.

2 - Détection et classification de changements sur des scènes urbaines en télédétection. A. Fournier. PhD Thesis, Institut Supérieur de l'Aéronautique et de l'Espace, October 2008. Keywords : détection de changements, Satellite images, lignes de niveau, Classification, Urban areas, statistiques directionnelles. @PHDTHESIS{Fournier08, author = {Fournier, A.}, title = {Détection et classification de changements sur des scènes urbaines en télédétection}, year = {2008}, month = {October}, school = {Institut Supérieur de l'Aéronautique et de l'Espace}, url = {http://tel.archives-ouvertes.fr/tel-00463593/fr/}, keyword = {détection de changements, Satellite images, lignes de niveau, Classification, Urban areas, statistiques directionnelles} } Résumé : Cette thèse aborde le problème de la détection de changements sur des images de scènes urbaines en télédétection. Les expériences ont été menées sur des couples d'images satellitaires panchromatiques haute résolution (< 1 m). À travers ce thème général, plusieurs problématiques, correspondant aux divers niveaux d'une chaîne de traitement, sont abordés, depuis la création d'un masque de changements jusqu'au raisonnement à un niveau objet. Dans ce manuscrit, nous abordons premièrement le problème de la détermination d'un masque de changements. Après avoir étudié les limites d'un algorithme de détection de changements, fondé sur l'analyse en composantes principales, nous proposons un algorithme tirant parti de l'invariance des lignes de niveau, fondé sur un modèle d'illumination et des hypothèses sur la régularité de la scène. Par la suite, nous abordons la classification des zones détectées comme changées au cours de l'étape précédente. D'abord, nous nous fondons uniquement sur les radiométries des couples de pixels. Enfin, nous étudions l'intérêt d'une composante géométrique dans la classification. Plus précisément, nous appliquons un algorithme d'approximation polygonale sur les zones connexes issues de la classification précédentes, puis nous classifions les formes obtenues compte tenu des orientations des côtés des polygones obtenus. Abstract : This thesis addresses the problem of change detection on remotely sensed urban scenes. experiences were run on couples of high resolution (<1m) panchromatic satellite images. Through this general theme, different problems, corresponding to different levels of a processing chain were addressed, from the determination of a change mask to an object level reasoning. In this work, we first address the problem of determining a change mask. We study the assets and limits of a change detection algorithm based on a Principal Component Analysis. We then propose a new algorithm that relies on the invariance of the level lines. It is based on a simple illumination model and some hypotheses on the scene regularity. Then we address the classification of the zones detected as changed during our first step. This is done by only considering the radiometries of each pixel couple. Finally, we study the interest of a geometric component in our classification. More precisely, we apply a polygonal approximation algorithm on the connected zones generated by the first classification, then we classify the obtained shapes according to the orientations of the polygon edges.

3 - Détection de zones brûlées après un feu de forêt à partir d'une seule image satellitaire SPOT 5 par techniques SVM. O. Zammit. PhD Thesis, Universite de Nice Sophia Antipolis, September 2008. Keywords : Classification, Satellite images, Burnt areas, Forest fires, Support Vector Machines, Region Growing. Copyright : @PHDTHESIS{zammit_these_08, author = {Zammit, O.}, title = {Détection de zones brûlées après un feu de forêt à partir d'une seule image satellitaire SPOT 5 par techniques SVM}, year = {2008}, month = {September}, school = {Universite de Nice Sophia Antipolis}, url = {http://tel.archives-ouvertes.fr/tel-00345683/fr/}, keyword = {Classification, Satellite images, Burnt areas, Forest fires, Support Vector Machines, Region Growing} } Résumé : Cette thèse aborde le problème de cartographie de zones brûlées à partir d'images satellitaires haute résolution. Nos modèles reposent sur le traitement d'une seule image SPOT 5, acquise après le feu afin de détecter automatiquement les zones brûlées. Le modèle est fondé sur les Séparateurs à Vaste Marge (SVM), une technique de classification supervisée qui a démontré une meilleure précision et une meilleure capacité de généralisation que les algorithmes de classification plus traditionnels. Concernant notre problème de détection, les différentes zones brûlées possèdent des caractéristiques spectrales assez similaires, au contraire des zones non brûlées (végétation, routes, eau, zones urbaines, nuage, ombre...) dont les caractéristiques spectrales varient énormément. Nous proposons donc d'utiliser les One-Class SVM, une technique qui dérive des SVM mais qui n'utilise que des exemples de pixels brûlés pour les phases d'apprentissage et de classification. Afin de prendre en compte l'information spatiale de l'image, l'algorithme OC-SVM est utilisé comme une technique de croissance de régions, ce qui permet de diminuer les fausses alarmes et d'améliorer les contours des zones brûlées. De plus, la base d'exemple de pixels brûlés nécessaire à l'apprentissage des techniques SVM est déterminée automatiquement à partir de l'histogramme de l'image. Finalement, la méthode de classification proposée est testée sur plusieurs images satellitaires afin de valider son efficacité selon le type de végétation et la surface des zones brûlées. Les zones brûlées obtenues sont comparées aux vérités de terrain fournies par le CNES, Infoterra France, le SERTIT, les Services Départementaux d'Incendies et de Secours ou l'Office National des Forêts.

4 - Indexing of satellite images using structural information. A. Bhattacharya. PhD Thesis, Ecole Nationale Supérieure des Télécommunications, 2007. Keywords : Landscape, Segmentation, Features, Extraction, Classification, Data mining. @PHDTHESIS{bhattacharya_these, author = {Bhattacharya, A.}, title = {Indexing of satellite images using structural information}, year = {2007}, school = {Ecole Nationale Supérieure des Télécommunications}, pdf = {ftp://ftp-sop.inria.fr/ariana/Articles/2007_bhattacharya_these.pdf}, keyword = {Landscape, Segmentation, Features, Extraction, Classification, Data mining} }

5 - Contribution à l'Analyse de Textures en Traitement d'Images par Méthodes Variationnelles et Equations aux Dérivées Partielles. J.F. Aujol. PhD Thesis, Universite de Nice Sophia Antipolis, June 2004. Keywords : Image decomposition, Classification, Restoration, Fonctional analysis, Bounded Variation Space, Sobolev space. @PHDTHESIS{JFAujol, author = {Aujol, J.F.}, title = {Contribution à l'Analyse de Textures en Traitement d'Images par Méthodes Variationnelles et Equations aux Dérivées Partielles}, year = {2004}, month = {June}, school = {Universite de Nice Sophia Antipolis}, url = {https://hal.inria.fr/tel-00006303}, pdf = {http://hal.inria.fr/docs/00/04/68/89/PDF/tel-00006303.pdf}, keyword = {Image decomposition, Classification, Restoration, Fonctional analysis, Bounded Variation Space, Sobolev space} } Résumé : Cette thèse est un travail en mathématiques appliquées. Elle aborde quelques problèmes en analyse d'images et utilise des outils mathématiques spécifiques. L'objectif des deux premières parties de cette thèse est de proposer un modèle pour décomposer une image f'en trois composantes : f=u+v+w. Notre approche repose sur l'utilisation d'espaces mathématiques adaptés à chaque composante: l'espace BV des fonctions à variations bornées pour u, un espace G'proche du dual de BV pour les textures, et un espace de Besov d'exposant négatif E'pour le bruit. Nous effectuons l'étude mathématique complète des différents modèles que nous proposons. Nous illustrons notre approche par de nombreux exemples.Dans la troisième et dernière partie de cette thèse, nous nous intéressons spécifiquement à la composante texturée. Nous proposons un algorithme de classification supervisée pour les images texturées. Abstract : This Ph.D. thesis is a work in applied mathematics. It deals with image processing problems, and uses specific mathematical tools. The aim of the two first parts is to propose a model for decomposing an image f'into three components : f=u+v+w. Our approach relies on the use of mathematical spaces adapted to each component : the space BV of functions with bounded variations for u, a space G'close to the dual space of BV for v, and a negative Besov space E'for w. We carry out the complete mathematical analysis of the different models we propose. We illustrate our approach with many numerical examples. In the third and last part, we only deal with the texture component of an image. We propose a supervised classification algorithm for textured images.

6 - Analyse de texture dans l'espace hyperspectral par des méthodes probabilistes. G. Rellier. PhD Thesis, Universite de Nice Sophia Antipolis, November 2002. Keywords : Hyperspectral imaging, Texture, Classification, Markov Fields. @PHDTHESIS{rellier, author = {Rellier, G.}, title = {Analyse de texture dans l'espace hyperspectral par des méthodes probabilistes}, year = {2002}, month = {November}, school = {Universite de Nice Sophia Antipolis}, url = {https://hal.inria.fr/tel-00505898}, keyword = {Hyperspectral imaging, Texture, Classification, Markov Fields} } Résumé : Dans cette thèse, on aborde le problème de l'analyse de texture pour l'étude des zones urbaines. La texture est une notion spatiale désignant ce qui, en dehors de la couleur ou du niveau de gris, caractérise l'homogénéité visuelle d'une zone donnée d'une image. Le but de cette étude est d'établir un modèle qui permette une analyse de texture prenant en compte conjointement l'aspect spatial et l'aspect spectral, à partir d'images hyperspectrales. Ces images sont caractérisées par un nombre de canaux largement supérieur à celui des images multispectrales classiques. On désire tirer parti de l'information spectrale pour améliorer l'analyse spatiale. Les textures sont modélisées par un champ de Markov gaussien vectoriel, qui permet de prendre en compte les relations spatiales entre pixels, mais aussi les relations inter-bandes à l'intérieur d'un même pixel. Ce champ est adapté aux images hyperspectrales par une simplification évitant l'apparition de problèmes d'estimation statistique dans des espaces de grande dimension. Dans le but d'éviter ces problèmes, on effectue également une réduction de dimension des données grâce à un algorithme de poursuite de projection. Cet algorithme permet de déterminer un sous-espace de projection dans lequel une grandeur appelée indice de projection est optimisée. L'indice de projection est défini par rapport à la modélisation de texture proposée, de manière à ce que le sous-espace optimal maximise la distance entre les classes prédéfinies, dans le cadre de la classification. La méthode d'analyse de texture est testée dans le cadre d'une classification supervisée. Pour ce faire, on met au point deux algorithmes que l'on compare avec des algorithmes classiques utilisant ou non l'information de texture. Des tests sont réalisés sur des images hyperspectrales AVIRIS. Abstract : In this work, we investigate the problem of texture analysis of urban areas. Texture is a spatial concept that refers to the visual homogeneity characteristics of an image, not taking into account color or grey level. The aim of this research is to define a model which allows a joint spectral and spatial analysis of texture, and then to apply this model to hyperspectral images. These images many more bands than classical multispectral images. We intend to make use of spectral information and improve simple spatial analysis. Textures are modeled by a vectorial Gauss-Markov random field, which allows us to take into account the spatial interactions between pixels as well as inter-band relationships for a single pixel. This field has been adapted to hyperspectral images by a simplification which avoids statistical estimation problems common to high dimensional spaces. In order to avoid these problems, we also reduce the dimensionality of the data, using a projection pursuit algorithm. This algorithm determines a projection subspace in which an index, called projection index, is optimized. This index is defined in relation to the proposed texture model so that, when a classification is being carried out, the optimal subspace maximizes the distance between predefined training samples. This texture analysis method is tested within a supervised classification framework. For this purpose, we propose two classification algorithms that we compare to two classical algorithms, one which uses texture information and one which does not. Tests are carried out on AVIRIS hyperspectral images.

7 - Contribution à la classification d'images satellitaires par approche variationnelle et équations aux dérivées partielles. C. Samson. PhD Thesis, Universite de Nice Sophia Antipolis, September 2000. Keywords : Classification, Restoration, Level sets, Active contour. @PHDTHESIS{cs, author = {Samson, C.}, title = {Contribution à la classification d'images satellitaires par approche variationnelle et équations aux dérivées partielles}, year = {2000}, month = {September}, school = {Universite de Nice Sophia Antipolis}, url = {https://tel.archives-ouvertes.fr/tel-00319709}, pdf = {http://tel.archives-ouvertes.fr/docs/00/31/97/09/PDF/SAMSONthesis.pdf}, keyword = {Classification, Restoration, Level sets, Active contour} } Résumé : Ce travail est consacré au développement ainsi qu'à l'implantation de deux modèles variationnels pour la classification d'images. La classification d'images, consistant à attribuer une étiquette à chaque pixel d'une image, concerne de nombreuses applications à partir du moment où cette opération intervient très souvent à la base des chaînes de traitement et d'interprétation d'images. De nombreux modèles de classification ont déjà été développés dans un cadre stochastique ou à travers des approches structurales, mais rarement dans un contexte variationnel qui a déjà montré son efficacité dans divers domaines tels que la reconstruction ou la restauration d'images. Le premier modèle que nous proposons repose sur la minimisation d'une famille de critères dont la suite de solutions converge vers une partition des données composée de classes homogènes séparées par des contours réguliers. Cette approche entre dans le cadre des problèmes à discontinuité libre (it free discontinuity problems) et fait appel à des notions de convergence variationnelle telle que la théorie de la Gamma-convergence. La famille de fonctionnelles que nous proposons de minimiser contient un terme de régularisation, ainsi qu'un terme de classification. Lors de la convergence de cette suite de critères, le modèle change progressivement de comportement en commençant par restaurer l'image avant d'entamer le processus d'étiquetage des pixels. Parallèlement à cette approche, nous avons développé un second modèle de classification mettant en jeu un ensemble de régions et contours actifs. Nous utilisons une approche par ensembles de niveaux pour définir le critère à minimiser, cette approche ayant déjà suscité de nombreux travaux dans le cadre de la segmentation d'images. Chaque classe, et son ensemble de régions et contours associé, est défini à travers une fonction d'ensemble de niveaux. Le critère contient des termes reliés à l'information sur les régions ainsi qu'à l'information sur les contours. Nous aboutissons à la résolution d'un système d'équations aux dérivées partielles couplées et plongées dans un schéma dynamique. L'évolution de chaque région est guidée par un jeu de forces permettant d'obtenir une partition de l'image composée de classes homogènes et dont les frontières sont lisses. Nous avons mené des expériences sur de nombreuses données synthétiques ainsi que sur des images satellitaires SPOT. Nous avons également étendu ces deux modèles au cas de données multispectrales et obtenu des résultats sur des données SPOT XS que nous avons comparé à ceux obtenus par différents modèles. Abstract : This work is devoted to the development and the implementation of variational models for image classification.\ Image classification, which consists in assiging a label to each pixel of a given image, concerns many applications since it is often the basic processing for many image interpretation systems. Many models have been developed within a stochastic framework or using structural approaches, but rarely within a variational framework whose efficiency has largely been proved for a wide variety of problems such as image reconstruction or restoration. The first model we propose herein is based on the minimization of a criterion family whose set of solutions in converging to a partition of the data set composed of homogeneous regions with regularized boundaries. This approach takes place within the context of free boundary problems and we use the Gamma-convergence theory for the theoretical study. The set of functionals we minimize contains a regularization term and a classification one. As the set of functionals is converging, the behavior of the model is progressively changing: the restoration process is vanishing while the labeling one is rising. The second model we propose is based on a set of active regions and contours. We use a level set formulation to define the criterion we want to minimize, this formulation allows a change of topology of the evolving sets. Each class and its associated set of regions and boundaries is defined thanks to a level set function. From the Euler equations, we solve a system of coupled partial differential equations through a dynamical scheme. The evolution of each region is governed by forces constraining the partition to be composed of homogeneous classes with smooth boundaries.\ We have conducted many experiments on both synthetic and real images. We have extended these models to the multispectral case for which the data are a set of images, and we show some results and comparisons on SPOT XS images.

top of the page

12 Conference articles

1 - Synthetic Aperture Radar Image Classification via Mixture Approaches. V. Krylov and J. Zerubia. In Proc. IEEE International Conference on Microwaves, Communications, Antennas and Electronic Systems (COMCAS), Tel Aviv, Israel, November 2011. Keywords : Synthetic Aperture Radar (SAR), remote sensing, high resolution, Classification, finite mixture models, generalized gamma distribution. Copyright : IEEE @INPROCEEDINGS{krylovCOMCAS11, author = {Krylov, V. and Zerubia, J.}, title = {Synthetic Aperture Radar Image Classification via Mixture Approaches}, year = {2011}, month = {November}, booktitle = {Proc. IEEE International Conference on Microwaves, Communications, Antennas and Electronic Systems (COMCAS)}, address = {Tel Aviv, Israel}, url = {http://www.ortra.biz/comcas/}, pdf = {http://hal.archives-ouvertes.fr/inria-00625551/en/}, keyword = {Synthetic Aperture Radar (SAR), remote sensing, high resolution, Classification, finite mixture models, generalized gamma distribution} } Abstract : In this paper we focus on the fundamental synthetic aperture radars (SAR) image processing problem of supervised classification. To address it we consider a statistical finite mixture approach to probability density function estimation. We develop a generalized approach to address the problem of mixture estimation and consider the use of several different classes of distributions as the base for mixture approaches. This allows performing the maximum likelihood classification which is then refined by Markov random field approach, and optimized by graph cuts. The developed method is experimentally validated on high resolution SAR imagery acquired by Cosmo-SkyMed and TerraSAR-X satellite sensors.

2 - SAR image classification with non- stationary multinomial logistic mixture of amplitude and texture densities. K. Kayabol and A. Voisin and J. Zerubia. In Proc. IEEE International Conference on Image Processing (ICIP), pages 173-176, Brussels, Belgium, September 2011. Keywords : High resolution SAR images, Classification, Texture, Multinomial logistic, Classification EM algorithm. @INPROCEEDINGS{inria-00592252, author = {Kayabol, K. and Voisin, A. and Zerubia, J.}, title = {SAR image classification with non- stationary multinomial logistic mixture of amplitude and texture densities}, year = {2011}, month = {September}, booktitle = {Proc. IEEE International Conference on Image Processing (ICIP)}, pages = {173-176}, address = {Brussels, Belgium}, url = {http://hal.inria.fr/inria-00592252/en/}, keyword = {High resolution SAR images, Classification, Texture, Multinomial logistic, Classification EM algorithm} } Abstract : We combine both amplitude and texture statistics of the Synthetic Aperture Radar (SAR) images using Products of Experts (PoE) approach for classification purpose. We use Nakagami density to model the class amplitudes. To model the textures of the classes, we exploit a non-Gaussian Markov Random Field (MRF) texture model with t-distributed regression error. Non-stationary Multinomial Logistic (MnL) latent class label model is used as a mixture density to obtain spatially smooth class segments. We perform the classification Expectation-Maximization (CEM) algorithm to estimate the class parameters and classify the pixels. We obtained some classification results of water, land and urban areas in both supervised and semi-supervised cases on TerraSAR-X data.

3 - Classification bayésienne supervisée d’images RSO de zones urbaines à très haute résolution. A. Voisin and V. Krylov and J. Zerubia. In Proc. GRETSI Symposium on Signal and Image Processing, Bordeaux, September 2011. Keywords : SAR Images, Classification, Urban areas, Markov Fields, Hierarchical models. @INPROCEEDINGS{VoisinGretsi2011, author = {Voisin, A. and Krylov, V. and Zerubia, J.}, title = {Classification bayésienne supervisée d’images RSO de zones urbaines à très haute résolution}, year = {2011}, month = {September}, booktitle = {Proc. GRETSI Symposium on Signal and Image Processing}, address = {Bordeaux}, url = {http://hal.inria.fr/inria-00623003/fr/}, keyword = {SAR Images, Classification, Urban areas, Markov Fields, Hierarchical models} } Résumé : Ce papier présente un modèle de classification bayésienne supervisée d’images acquises par Radar à Synthèse d’Ouverture (RSO) très haute résolution en polarisation simple contenant des zones urbaines, particulièrement affectées par le bruit de chatoiement. Ce modèle prend en compte à la fois une représentation statistique des images RSO par modèle de mélanges finis et de copules, et une modélisation contextuelle à partir de champs de Markov hiérarchiques. Abstract : This paper deals with the Bayesian classification of single-polarized very high resolution synthetic aperture radar (SAR) images that depict urban areas. The difficulty of such a classification relies in the significant effects of speckle noise. The model considered here takes into account both statistical modeling of images via finite mixture models and copulas, and contextual modeling thanks to hierarchical Markov random fields

4 - Morphological road segmentation in urban areas from high resolution satellite images. R. Gaetano and J. Zerubia and G. Scarpa and G. Poggi. In International Conference on Digital Signal Processing, Corfu, Greece, July 2011. Keywords : Segmentation, Classification, skeletonization , pattern recognition, shape analysis. @INPROCEEDINGS{GaetanoDSP, author = {Gaetano, R. and Zerubia, J. and Scarpa, G. and Poggi, G.}, title = {Morphological road segmentation in urban areas from high resolution satellite images}, year = {2011}, month = {July}, booktitle = {International Conference on Digital Signal Processing}, address = {Corfu, Greece}, url = {http://hal.inria.fr/inria-00618222/fr/}, keyword = {Segmentation, Classification, skeletonization , pattern recognition, shape analysis} } Abstract : High resolution satellite images provided by the last generation sensors significantly increased the potential of almost all the image information mining (IIM) applications related to earth observation. This is especially true for the extraction of road information, task of primary interest for many remote sensing applications, which scope is more and more extended to complex urban scenarios thanks to the availability of highly detailed images. This context is particularly challenging due to such factors as the variability of road visual appearence and the occlusions from entities like trees, cars and shadows. On the other hand, the peculiar geometry and morphology of man-made structures, particularly relevant in urban areas, is enhanced in high resolution images, making this kind of information especially useful for road detection. In this work, we provide a new insight on the use of morphological image analysis for road extraction in complex urban scenarios, and propose a technique for road segmentation that only relies on this domain. The keypoint of the technique is the use of skeletons as powerful descriptors for road objects: the proposed method is based on an ad-hoc skeletonization procedure that enhances the linear structure of road segments, and extracts road objects by first detecting their skeletons and then associating each of them with a region of the image. Experimental results are presented on two different high resolution satellite images of urban areas.

5 - Multichannel SAR Image Classification by Finite Mixtures, Copula Theory and Markov Random Fields. V. Krylov and G. Moser and S.B. Serpico and J. Zerubia. In Proc. of Bayesian Inference and Maximum Entropy Methods in Science and Engineering (MaxEnt 2010), Vol. 1305, pages 319-326, Chamonix, France, July 2010. Keywords : multichannel SAR, Classification, probability density function estimation, Markov random field, copula. Copyright : AIP @INPROCEEDINGS{krylovMaxEnt10, author = {Krylov, V. and Moser, G. and Serpico, S.B. and Zerubia, J.}, title = {Multichannel SAR Image Classification by Finite Mixtures, Copula Theory and Markov Random Fields}, year = {2010}, month = {July}, booktitle = {Proc. of Bayesian Inference and Maximum Entropy Methods in Science and Engineering (MaxEnt 2010)}, volume = {1305}, pages = {319-326}, address = {Chamonix, France}, url = {http://hal.archives-ouvertes.fr/inria-00495557/en/}, pdf = {http://hal.archives-ouvertes.fr/docs/00/49/55/57/PDF/krylov_MaxEnt2010.pdf}, keyword = {multichannel SAR, Classification, probability density function estimation, Markov random field, copula} } Abstract : The last decades have witnessed an intensive development and a significant increase of interest to remote sensing, and, in particular, to synthetic aperture radar (SAR) imagery. In this paper we develop a supervised classification approach for medium and high resolution multichannel SAR amplitude images. The proposed technique combines finite mixture modeling for probability density function estimation, copulas for multivariate distribution modeling and the Markov random field approach to Bayesian image classification. The finite mixture modeling is done via a recently proposed SAR-specific dictionary-based stochastic expectation maximization approach to class-conditional amplitude probability density function estimation, which is applied separately to all the SAR channels. For modeling the class-conditional joint distributions of multichannel data the statistical concept of copulas is employed, and a dictionary-based copula selection method is proposed. Finally, the Markov random field approach enables to take into account the contextual information and to gain robustness against the inherent noise-like phenomenon of SAR known as speckle. The designed method is an extension and a generalization to multichannel SAR of a recently developed single-channel and Dual-pol SAR image classification technique. The accuracy of the developed multichannel SAR classification approach is validated on several multichannel Quad-pol RADARSAT-2 images and compared to benchmark classification techniques.

6 - Unsupervised One-Class SVM Using a Watershed Algorithm and Hysteresis Thresholding to Detect Burnt Areas. O. Zammit and X. Descombes and J. Zerubia. In Proc. International Conference on Pattern Recognition and Image Analysis (PRIA), Nizhny Novgorod, Russia, September 2008. Keywords : Classification, Segmentation, Support Vector Machines, Burnt areas, Forest fires, Satellite images. Copyright : @INPROCEEDINGS{zammit_pria_08, author = {Zammit, O. and Descombes, X. and Zerubia, J.}, title = {Unsupervised One-Class SVM Using a Watershed Algorithm and Hysteresis Thresholding to Detect Burnt Areas}, year = {2008}, month = {September}, booktitle = {Proc. International Conference on Pattern Recognition and Image Analysis (PRIA)}, address = {Nizhny Novgorod, Russia}, pdf = {http://hal.inria.fr/inria-00316297/fr/}, keyword = {Classification, Segmentation, Support Vector Machines, Burnt areas, Forest fires, Satellite images} }

7 - Combining One-Class Support Vector Machines and hysteresis thresholding: application to burnt area mapping. O. Zammit and X. Descombes and J. Zerubia. In Proc. European Signal Processing Conference (EUSIPCO), Lausanne, Switzerland, August 2008. Note : to appear. Keywords : Classification, Satellite images, Support Vector Machines, Burnt areas, Forest fires, Clustering. Copyright : @INPROCEEDINGS{zammit_eusipco_08, author = {Zammit, O. and Descombes, X. and Zerubia, J.}, title = {Combining One-Class Support Vector Machines and hysteresis thresholding: application to burnt area mapping}, year = {2008}, month = {August}, booktitle = {Proc. European Signal Processing Conference (EUSIPCO)}, address = {Lausanne, Switzerland}, url = {http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=7080254}, keyword = {Classification, Satellite images, Support Vector Machines, Burnt areas, Forest fires, Clustering} }

8 - Indexing of mid-resolution satellite images with structural attributes. A. Bhattacharya and M. Roux and H. Maitre and I. H. Jermyn and X. Descombes and J. Zerubia. In The International Society for Photogrammetry and Remote Sensing, Beijing, China, July 2008. Keywords : Landscape, Segmentation, Features, Extraction, Classification, Modelling. @INPROCEEDINGS{Bhattacharya08, author = {Bhattacharya, A. and Roux, M. and Maitre, H. and Jermyn, I. H. and Descombes, X. and Zerubia, J.}, title = {Indexing of mid-resolution satellite images with structural attributes}, year = {2008}, month = {July}, booktitle = {The International Society for Photogrammetry and Remote Sensing}, address = {Beijing, China}, pdf = {http://www-sop.inria.fr/members/Ian.Jermyn/publications/Bhattacharya08isprs.pdf}, keyword = {Landscape, Segmentation, Features, Extraction, Classification, Modelling} } Abstract : Indexing and retrieval of satellite images relies on the extraction of appropriate information from the data about the entity of interest (e.g. land cover type) and on the robustness of this extraction to nuisance variables. Entities in an image may be strongly correlated with each other and can therefore be used to characterize geographical environments on the Earth’s surface. The properties of road networks vary considerably from one geographical environment to another. The networks pertaining in a satellite image can therefore be used to classify and retrieve such environments. In the work presented in this paper we have defined 7 such classes. These classes can be categorized as follows: 2 urban classes consisting of “Urban USA” and “Urban Europe”; 3 rural classes consisting of “Villages”, “Mountains” and “Fields”; an “Airports” class and a “Common” class (this can be considered as a rejection class). These classes were then classified with the aid of geometrical and topological features computed from the road networks occurring in them. In our work we have used two extraction methods simultaneously on an image to extract the road networks pertaining in it. A set of 16 network features were computed from one extraction method and were categorized into 6 groups as follows: 6 measures of ‘density’, 4 measures of ‘curviness’, 2 measures of ‘homogeneity’, 1 measure of ‘length’, 2 measures of ‘distribution’ and 1 measure of ‘entropy’. Due to certain limitations of these extraction methods there was a relative failure of network extraction in certain urban regions con- taining narrow and dense road structures. This loss of information was circumvented by segmenting the urban regions and computing a second set of geometrical and topological features from them. A set of 4 urban region features were computed and were categorized into 3 groups as follows: 2 measures of ‘density’, 1 measure of ‘labels’ and 1 measure of ‘compactness’. The 500 images (each of size 512x512 pixels) forming our database were selected from SPOT5 scenes with 5m resolution. From each image a set of geometrical and topological features were computed from the road networks and urban regions. These features were then used to classify the pre-defined geographical classes. Feature selection was done to avoid the burden of feature dimensionality and increase the classification performance. A set of 20 features was selected from 36 features by Fisher Linear Discriminant (FLD) analysis which gave the least classification error with an one-vs-rest linear Support Vector Machine (SVM). The impact of spatial resolution and size of images on the feature set have been explored in this work. We took a closer look at the effect of spatial resolution and size of images on the discriminative power of the feature set to classify the images belonging to the pre-defined geographical classes. Tests were performed with feature selection by FLD and one-vs-rest linear SVM classification on a database with images of 10m resolution. Another test was performed with feature selection by FLD and one-vs-rest linear SVM classification on a database with 5m resolution images (each of size 256x256 pixels). With the above mentioned approaches, we developed a novel method to classify large satellite images acquired by SPOT5 satellite (5m resolution) with patches of images each of size 512x512 pixels extracted from them. There has been a large amount of work dedicated to the classification of large satellite images at pixel level rather than considering image patches of different sizes. Classification of image patches of different sizes from a large satellite image is a novel idea in the sense that the patches considered contain significant coverage of a particular type of geographical environment. Road networks and urban region features were computed from these image patches extracted from the large image. A one-vs-rest Gaussian kernel SVM classification method was used to classify this large image. The classification results show that the image patches were labeled with the class having the maximum geographical coverage of the area associated in the large image. The large image was mapped into a “region matrix”, where each element of the matrix corresponds to a geographical class. This is a ‘hard’ classification and no inference can be drawn about the classification confidence. In certain cases, this produces some anomalies, as a single patch may contain two or more different geographical coverages. In order to have an estimate of these partial coverages, the output of the SVM was mapped into probabilities. These probability measures were then studied to have a closer look at the classification accuracies. The results confirm that our method is able to classify a large image into various geographical classes with a mean error of less than 10%. Future studies can use operators to detect not only man-made structures like roads and urban areas, but also natural entities like rivers, forests, etc. In this work we have restricted ourselves to a single resolution, but our methodology can be adapted to consider images of higher resolutions from QuickBird and the future Pleiade satellite. At a better resolution it may be possible to extract different structures like buildings, gardens, cross-roads, etc. This in turn will allow us to incorporate more classes to appropriately classify any geographical environment. At an image resolution of 1m, we may imagine to have sub-classes of an existing class, e.g., classes like urban Europe and urban USA can de divided into downtown, residential and industrial classes.

9 - Mixing Geometric and Radiometric Features for Change Classification. A. Fournier and X. Descombes and J. Zerubia. In Proc. SPIE Symposium on Electronic Imaging, San Jose, USA, January 2008. Keywords : Change detection, directional Statistics, polygonal approximation, Classification. Copyright : Copyright 2008 SPIE and IS&T. This paper was published in the proceedings of IS&T/SPIE 20th Annual Symposium on Electronic Imaging and is made available as an electronic reprint (preprint) with permission of SPIE and IS&T. One print or electronic copy may be made for personal use only. Systematic or multiple reproduction, distribution to multiple locations via electronic or other means, duplication of any material in this paper for a fee or for commercial purposes, or modification of the content of the paper are prohibited. @INPROCEEDINGS{fournier_spie08, author = {Fournier, A. and Descombes, X. and Zerubia, J.}, title = {Mixing Geometric and Radiometric Features for Change Classification}, year = {2008}, month = {January}, booktitle = {Proc. SPIE Symposium on Electronic Imaging}, address = {San Jose, USA}, url = {http://hal.inria.fr/inria-00269853/fr/}, keyword = {Change detection, directional Statistics, polygonal approximation, Classification} } Abstract : Most basic change detection algorithms use a pixel-based approach. Whereas such approach is quite well defined for monitoring important area changes (such as urban growth monitoring) in low resolution images, an object based approach seems more relevant when the change detection is specifically aimed toward targets (such as small buildings and vehicles). In this paper, we present an approach that mixes radiometric and geometric features to qualify the changed zones. The goal is to establish bounds (appearance, disappearance, substitution ...) between the detected changes and the underlying objects. We proceed by first clustering the change map (containing each pixel bitemporal radiosity) in different classes using the entropy-kmeans algorithm. Assuming that most man-made objects have a polygonal shape, a polygonal approximation algorithm is then used in order to characterize the resulting zone shapes. Hence allowing us to refine the primary rough classification, by integrating the polygon orientations in the state space. Tests are currently conducted on Quickbird data.

10 - Apprentissage non supervisé des SVM par un algorithme des K-moyennes entropique pour la détection de zones brûlées. O. Zammit and X. Descombes and J. Zerubia. In Proc. GRETSI Symposium on Signal and Image Processing, Troyes, France, September 2007. Keywords : Satellite images, Forest fires, Burnt areas, Classification, Support Vector Machines, Learning base. @INPROCEEDINGS{zammit_gretsi_07, author = {Zammit, O. and Descombes, X. and Zerubia, J.}, title = {Apprentissage non supervisé des SVM par un algorithme des K-moyennes entropique pour la détection de zones brûlées}, year = {2007}, month = {September}, booktitle = {Proc. GRETSI Symposium on Signal and Image Processing}, address = {Troyes, France}, pdf = {ftp://ftp-sop.inria.fr/ariana/Articles/2007_zammit_gretsi_07.pdf}, keyword = {Satellite images, Forest fires, Burnt areas, Classification, Support Vector Machines, Learning base} }

top of the page

navigation : 1 - 2 display : 10 per page | 20 per page | 50 per page

These pages were generated by