Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Cet article presente un nouveau modele de codage pour la classification a partir d’un ensemble de matrices de covariance : la matrice de co-occurrence associee a un dictionnaire de matrices de covariance. Contrairement aux modeles de... more
Cet article presente un nouveau modele de codage pour la classification a partir d’un ensemble de matrices de covariance : la matrice de co-occurrence associee a un dictionnaire de matrices de covariance. Contrairement aux modeles de codage de l'etat de l'art (SDMR, R-VLAD and VRF), un tel modele local exploite la repartition spatiale des patchs. A partir du modele de melange generatif de distributions gaussiennes Riemanniennes, nous presentons ce modele local. Une experience sur la classification d'images de texture est ensuite effectuee sur les bases de textures VisTex et Outex_TC000_13 afin de evaluer le potentiel de la methode proposee.
Research Interests:
Le probleme de pansharpening se situe dans la conception d'un procede permettant de fusionner une image pan-chromatique haute resolution avec une image multispectrale basse resolution an de generer une image multispectrale haute... more
Le probleme de pansharpening se situe dans la conception d'un procede permettant de fusionner une image pan-chromatique haute resolution avec une image multispectrale basse resolution an de generer une image multispectrale haute resolution. Dans la cadre de la recherche d'une solution par approche regularisee, nous etudions, dans cette communication, un terme d'attache aux donnees non local pour le probleme de pansharpening fonde sur un terme exploitant la geometrie de l'image. Ce terme est base sur l'hypothese que l'on peut aligner le gradient de l'image multispectrale haute resolution que l'on cherche en chaque point en regardant la direction des vecteurs donnes par le gradient de l'image panchromatique au voisinage de ce point. Nous explicitons le gradient de ce terme et eectuons des tests sur des donnees simulees et des donnees satellites.
Research Interests:
Most of the recent top-ranked stereovision algorithms are based on some oversegmentation of the stereo pair images. Each image segment is modelled as a 3D plane, transposing the stereovision problem into a 3-parameters plane estimation... more
Most of the recent top-ranked stereovision algorithms are based on some oversegmentation of the stereo pair images. Each image segment is modelled as a 3D plane, transposing the stereovision problem into a 3-parameters plane estimation problem. Trying to find the set of planes that best fits the input data while regularising on 3D adjacent planes, a global energy is minimised using different optimisation strategies. The regularisation term is important as it enforces the reconstructed scene consistency. In this paper, we propose a disparity map refinement strategy based on a triangular mesh of the image domain. This mesh is projected along the rays of the camera, each triangle being modelled by a slanted plane. Considering triangles instead of free form superpixels, i.e. adding a geometrical constraint on the shape of segment boundaries, allows to formulate simpler regularisation terms that only involve the mesh vertex disparities in the computations. Different curvature regularisat...
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Publication in the conference proceedings of EUSIPCO, Aalborg, Denmark, 2010
Research Interests:
Publication in the conference proceedings of EUSIPCO, Bucharest, Romania, 2012
Research Interests:
Research Interests: Physics and Indexation
ABSTRACT Dans ce papier, nous nous intéressons à l'indexation d'images texturées fondée sur la modélisation stochastique multivarié. Celle-ci permet par exemple de caractériser les dépendances spatiale et couleur des... more
ABSTRACT Dans ce papier, nous nous intéressons à l'indexation d'images texturées fondée sur la modélisation stochastique multivarié. Celle-ci permet par exemple de caractériser les dépendances spatiale et couleur des coefficients d'une décomposition en ondelettes. La problématique de ces travaux concerne le choix du modèle stochastique à utiliser pour modéliser tel ou tel type de dépendance. L'originalité de ces travaux réside dans l'utilisation de la distance géodésique pour mesurer la similarité entre des variables aléatoires elliptiquement distribuées. Dans ce contexte, nous donnons une expression analytique de la distance géodésique pour une distribution G0 lorsque les paramètres de forme et d'échelle sont fixés et lorsque l'on approxime les coordonnées des géodésiques par des lignes droites. Par le biais d'une étude comparative sur la base de données VisTex, nous montrons que les distributions multivariées de Laplace et G0 sont les meilleurs modèles parmi ceux considérés pour prendre en compte respectivement les dépendances couleur et spatiale. A travers une approche multi-modèles, nous montrons également qu'il est préférable de modéliser séparément ces dépendances au lieu de travailler sur un vecteur joint qui les caractériserait simultanément.
Research Interests:
Publication in the conference proceedings of EUSIPCO, Toulouse, France, 2002
Research Interests:
Research Interests:
Research Interests:
Publication in the conference proceedings of EUSIPCO, Marrakech, Morocco, 2013
Research Interests:
Publication in the conference proceedings of EUSIPCO, Bucharest, Romania, 2012
Research Interests:
Publication in the conference proceedings of EUSIPCO, Marrakech, Morocco, 2013
Research Interests:
Abstract—Due to its heavy-tailed and fully parametric form, the multivariate generalized Gaussian distribution (MGGD) has been receiving much attention for modeling extreme events in signal and image processing applications. Considering... more
Abstract—Due to its heavy-tailed and fully parametric form, the multivariate generalized Gaussian distribution (MGGD) has been receiving much attention for modeling extreme events in signal and image processing applications. Considering the estimation issue of the MGGD parameters, the main contribution of this paper is to prove that the maximum likelihood estimator (MLE) of the scatter matrix exists and is unique up to a scalar factor, for a given shape parameter β ∈ (0, 1). Moreover, an estimation algorithm based on a Newton-Raphson recursion is proposed for computing the MLE of MGGD parameters. Various experiments conducted on synthetic and real data are presented to illustrate the theoretical derivations in terms of number of iterations and number of samples for different values of the shape parameter. The main conclusion of this work is that the parameters of MGGDs can be estimated using the maximum likelihood principle with good performance. Index Terms—Multivariate generalized...
Research Interests:
Research Interests:
Research Interests: Philosophy and GSI
Research Interests:
In this paper, we propose to investigate stochastic sinusoidal models in order to characterise quasi-periodic signals. Indeed, in comparison to the broadly used autoregressive (AR) models, a sinusoidal approach seems to be more efficient... more
In this paper, we propose to investigate stochastic sinusoidal models in order to characterise quasi-periodic signals. Indeed, in comparison to the broadly used autoregressive (AR) models, a sinusoidal approach seems to be more efficient to capture quasi-periodic feature. Using AR process as a model for the sine wave magnitudes makes it possible to track the frequential non-stationarity of the signal. The scheme we propose operates as follows: once the frequency components of the signal are obtained, estimating the magnitudes of each sine component of the model is performed by means of an Expectation-Maximisation (EM) algorithm based on Kalman smoothing. Results are provided on sleep spindle and speech.
Research Interests:
In this paper, we propose an approach to improve high-resolution frequency estimation for narrow-band planes. This approach is based on a signal preprocessing combined with a high-resolution method to increase the accuracy of frequency... more
In this paper, we propose an approach to improve high-resolution frequency estimation for narrow-band planes. This approach is based on a signal preprocessing combined with a high-resolution method to increase the accuracy of frequency estimation. The preprocessing step is a Subband Decomposition Based on the Hilbert Transform (SDBHT) [1] for one and two-dimensional signals. This improvement is achieved by using an empirical criterion to determine the number of waves of the signals derived from the SDBHT technique. Simulation examples show the performances of this criterion. Then, we apply SDBHT method and empirical criterion to radar imaging.
Research Interests:
La plupart des methodes superpixels calculent un compromis entre des descripteurs spatiaux et couleur a l’echelle pixellique. Elles necessitent donc un reglage fin pour equilibrer ces mesures, et ne peuvent capturer une information de... more
La plupart des methodes superpixels calculent un compromis entre des descripteurs spatiaux et couleur a l’echelle pixellique. Elles necessitent donc un reglage fin pour equilibrer ces mesures, et ne peuvent capturer une information de texture. Dans ce travail, nous repondons a ces problemes avec une nouvelle methode robuste aux textures. Pour capturer les zones texturees et homogenes, la contrainte spatiale est ajustee automatiquement en fonction de la variance locale. Ensuite, pour assurer l’homogeneite de texture des superpixels, une nouvelle distance basee patchs est introduite. La methode proposee ameliore la precision de celles de l’etat de l’art sur des bases d’images texturees et naturelles couleur.
Research Interests:
In this paper, we investigate the problem of three-dimensional (3D) frequency estimation. We propose a new approach based on the shift invariance property in the data structure. The data are modeled as a sum of 3D complex exponential... more
In this paper, we investigate the problem of three-dimensional (3D) frequency estimation. We propose a new approach based on the shift invariance property in the data structure. The data are modeled as a sum of 3D complex exponential (SCE) embedded in white noise. In 1 and 2D cases, the approaches based on invariance property have shown efficiency, the purpose of this paper is to take advantage of this feature in the 3D framework. Indeed the special structure of the model permits a decomposition of the autocorrelation matrix into a linear subspace called signal subspace and its orthogonal complement, the noise subspace. The method operates in two steps, firstly one estimates the autocorrelation matrix which is defined and performed from a subset of data. Secondly the estimation of the frequencies is involved by the existence of an invertible matrix mapping between the signal subspace basis and an exact 3D Vandermonde matrix.
Research Interests:
Research Interests:
Despite the rapid development of computational hardware, the treatment of large and high dimensional data sets is still a challenging problem. The contribution of this paper to the topic is twofold. First, we propose a Gaussian mixture... more
Despite the rapid development of computational hardware, the treatment of large and high dimensional data sets is still a challenging problem. The contribution of this paper to the topic is twofold. First, we propose a Gaussian mixture model in conjunction with a reduction of the dimensionality of the data in each component of the model by principal component analysis, which we call PCA-GMM. To learn the (low dimensional) parameters of the mixture model we propose an EM algorithm whose M-step requires the solution of constrained optimization problems. Fortunately, these constrained problems do not depend on the usually large number of samples and can be solved efficiently by an (inertial) proximal alternating linearized minimization algorithm. Second, we apply our PCA-GMM for the superresolution of 2D and 3D material images based on the approach of Sandeep and Jacob. Numerical results confirm the moderate influence of the dimensionality reduction on the overall superresolution result.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Remote sensing image scene classification, which consists of labeling remote sensing images with a set of categories based on their content, has received remarkable attention for many applications such as land use mapping. Standard... more
Remote sensing image scene classification, which consists of labeling remote sensing images with a set of categories based on their content, has received remarkable attention for many applications such as land use mapping. Standard approaches are based on the multi-layer representation of first-order convolutional neural network (CNN) features. However, second-order CNNs have recently been shown to outperform traditional first-order CNNs for many computer vision tasks. Hence, the aim of this paper is to show the use of second-order statistics of CNN features for remote sensing scene classification. This takes the form of covariance matrices computed locally or globally on the output of a CNN. However, these datapoints do not lie in an Euclidean space but a Riemannian manifold. To manipulate them, Euclidean tools are not adapted. Other metrics should be considered such as the log-Euclidean one. This consists of projecting the set of covariance matrices on a tangent space defined at a...