Search

International audience

International audience

International audience; To discriminate gray-level texture images, spatial texture descriptors can be extracted using the local binary pattern (LBP) operator. In this paper we design an LBP operator that jointly extracts the spatial and spectral texture information directly from a raw image provided by a camera equipped with a multispectral filter array. Extensive experiments on a large dataset show that the proposed descriptor has both low computation cost and high discriminative power with regard to classical LBP descriptors applied to multispectral images.; Les motifs locaux binaires (LBP) permettent d'extraire des descripteurs de texture spatiaux pour discriminer les images de textures en niveaux de gris. Dans cet article, nous proposons un opérateur LBP qui extrait conjointement les informations de texture spatiales et spectrales à partir de l'image brute fournie par une caméra équipée d'une matrice de filtres multispectraux. Nous montrons expérimentalement que le descripteur proposé présente à la fois un faible coût de calcul et un haut pouvoir discriminant par rapport aux descripteurs LBP classiques appliqués aux images multispectrales.

International audience; Co-occurrence matrices (CMs) are well-known and efficient texture descriptors and fuzzy CMs (FCMs) have been designed to characterize grey-level texture images. In this paper, we propose to extend FCMs to color images and to apply them for efficient texture classification.

This chapter deals with the acquisition of colour images by mono-CCD colour cameras. These devices acquire only one colour component at each pixel through the CFA (Colour Filter Array) that covers the CCD sensor. A procedure - called demosaicing - is necessary to estimate the other two missing colour components at each pixel to obtain a colour image. A mathematical formalization for demosaicing is proposed, before we present some of the methods in the demosaicing literature, as well as the post-processing algorithms to correct the estimated images. We then present objective criteria for the quality evaluation of colour images, and apply them to estimated images to draw up conclusions about the selection of a demosaicing method. At last, we examine why and how white balancing is required and implemented to provide images faithfull to the observed scene under varying illumination conditions.

Within the framework of communicative gestures, sign language is the focus of our study, on account of the richness precisely awarded by its status of real language. We intend to elaborate a synthesis system of signed sentences from an intermediate textual representation that could be included in the global translation of written French into signs. From linguistically salient features, a formal grammar is proposed to specify signs, with the underlying principle of decomposition into formational primitives (hand configurations, movement, ...). The resulting hierarchic structure includes spatio-temporal features (body points, symmetry of the articulators, repetition) ; a specific study of speed profiles has lead besides to a precise model of sign dynamics. At the sentence level occur grammatical processes proper to getural languages : parametric signs allow us, in addition to describe generic lexical items, to overtake the mechanisms of localization and of pronominal reference. To mark clauses types, non-manual – especially facial – expressiveness is of major importance. The system has been implemented as a whole to result in the animation of a virtual signer. Obtaining natural configurations for the joints chains has required to develop a realistic model for the avatar, and dedicated inverse kinematics methods to orient and locate the hands. From the parser to the 3D generation module, the prototype is endowed with a graphical interface. As proves an illustrative example, it paves the way to a wide range of applications for which video is not suitable, taking advantage of the compact encoding and of the signs generation swiftness., Dans le cadre du geste communicatif, la langue des signes française (LSF) constitue un sujet d'étude privilégié, par la richesse que lui confère précisément son statut de langue. Notre étude vise à obtenir un système de synthèse de phrases en LSF à partir d'une représentation textuelle intermédiaire, dans l'optique d'une traduction globale du français en signes. Issue des traits linguistiquement pertinents, une grammaire formelle est proposée pour spécifier le signe, avec pour principe la décomposition en primitives formationnelles (configurations manuelles, mouvement, ...). La description hiérarchique obtenue inclut des caractéristiques spatio-temporelles (points corporels, symétrie des articulateurs, répétition); une étude poussée des profils de vitesse a par ailleurs été effectuée pour représenter finement la dynamique du mouvement. Au niveau du discours interviennent les processus grammaticaux propres aux langues gestuelles : le paramétrage des signes permet, outre de décrire des items lexicaux génériques, de prendre en charge les mécanismes de localisation ou de référence pronominale. Afin de marquer le type de clause, l'expressivité non-manuelle – notamment faciale – est de première importance. Le système a été intégralement implanté pour aboutir à l'animation d'un signeur virtuel. L'exigence de configurations naturelles pour les chaînes articulaires a nécessité le développement d'un modèle réaliste pour l'avatar, et de méthodes spécifiques de cinématique inverse pour l'orientation et le positionnement manuels. L'ensemble, de l'analyseur syntaxique au module de génération graphique tridimensionnelle, constitue un prototype performant d'obtention de phrases signées. Doté d'une interface graphique, il laisse entrevoir (comme le prouve un exemple illustratif) toute une gamme d'applications pour lesquelles la vidéo n'est pas adaptée, tirant principalement profit de la compacité de l'encodage et de la rapidité avec laquelle sont produits les signes.

International audience; The aim of this study is to use a color image segmentation based on pixel classification, where pixels of the image are grouped into classes. To do this, we proceed by analyzing the histogram of the image. We seek to identify the modes of the histogram, namely the areas of color space with a high density of probability. To identify these modes, mathematical morphology techniques are used. The application of watersheds on the color histogram leads to an over partitioning of the color plane, hence the need to merge modes using a new structure named the modes adjacency graph.

International audience

International audience; Mono-CCD color cameras acquire only one color component at each pixel by means of their color filter array (CFA) covering the CCD sensor. To obtain a color image, a procedure - called demosaicing - is then necessary to estimate the other two missing color components at each pixel. This chapter deals with the quality of color images generated in such a way. We attempt to determine which demosaicing method provides the best results according to several comparison criteria, particularly for subsequent needs of color image analyses like edge detection.

Special needs of deaf people appeal henceforward to sign language synthesis. The system presented here is based on a hierarchical description of sign, trying to take the different grammatical processes into account. Stress is laid on hand configurations specification thanks to finger shapes primitives and hand global properties, and on location and orientation computation issues. We then expose the results achieved from the corresponding written form of signs, leading to their computer virtual animation.