Le laboratoire XLIM (site du Futuroscope, ) recherche un doctorant qui sera financé dans le cadre du projet ANR DigiPi (Digital Pigment). La thèse se déroulera dans un contexte pluri-disciplinaire en collaboration avec le muséum national d'histoire naturelle de Paris pour la production de surfaces de références, et dans un contexte international (CIE) pour la validation des outils de mesure. Le recrutement est au niveau international (en anglais).
Pour toute information supplémentaire, ou pour proposer une candidature veuillez contacter : email@example.com
Texture Features for Hyperspectral Image Analysis
Phd direction: Noël Richard and Christine Fernandez-Maloigne
Keywords: Hyperspectral, Texture, Feature, Metrology, Non-Linear Processing
Thanks to recent advances in hyperspectral imaging sensors, images are acquired in high levels of spectral and spatial resolution. Combined, these resolutions allow to reach the optical properties (visible range) or physical properties (non-visible range), explaining aspects of the observed scene. Due to this direct link to the physical content of scene/ objects, the use of hyperspectral imaging continues to increase in many fields, e.g., remote sensing (aerial and satellite images), industry (quality control by vision), medical (diagnosis help) and cultural heritage (non-invasive methodology for preservation/ valorisation). However, there exists no tools allowing to assess the spectral content of the acquired images under metrological constraints.
Many current works try to address the question of material surface appearance by means of goniometric acquisition in colour and/or spectral domain [1, 2]. However, only a few take interests in the non-uniformity assessment of the acquired surface. In addition, there exists no hyperspectral texture feature that is processed in a full-band manner, explaining aspects of a textured and colored surface. Existing approaches extract texture attributes from gray-level/ intensity images obtained from selected wavelengths or spectral channels [1, 2]. The problem is, by using marginal approaches through selecting just some channels, these approaches lose the interest of the hyperspectral acquisition.
The main goal of this PhD work is to develop the first texture features to assess non-uniformity aspects of an acquired scene/ surface and to validate their metrological properties. We will extend some approaches originally developed for color image processing [3, 4, 5] to the spectral domain. The searched texture features will be based on distance functions, such that accuracy and other metrological properties can be managed and controlled [6, 7]. In previous studies, we have identified the spectral KL pseudo-divergence as a suitable spectral distance/ similarity measure to preserve metrological properties of spectral processing [8, 9]. Thanks to these previous developments and the collaboration with other laboratories, the PhD work will produce foundations for spectral metrology of non-uniform/textured surfaces. Applications of the constructed foundations will be oriented toward the cultural heritage domain. Nevertheless, validations of usefulness will also be carried out for satellite and aerial images in order to compare obtained results to existing bibliography and international contests in remote sensing.
The PhD project takes place within the ANR-DigiPi program (Digital Pigment: From Colour in Cultural Heritage to Industrial Requirements in Spectral Metrology) led by XLIM laboratory and the National Museum of Natural History (MNHN) of Paris. MNHN is responsible for the construction of spectral references and in the production of optical models relative to these references. This project will be also developed in collaboration with the MuvApp project led by the Colour and Visual Computing Laboratory (Colorlab) from NTNU (Gjøvik, Norway). The MuvApp project itself is dedicated to material appearance using goniometric acquisition and is interested to embed the textured/ non-uniformity aspects. Project meetings and communications will be organized within this international working group. The Phd position will start between October and December 2017.
 M. Dalla-Mura, A. Villa, J. Benediktsson, J. Chanussot et L. Bruzzone, «Classification of hyperspectral images by using extended morphological attribute profiles and Independant Component Analysis,» IEEE Journal of Geoscience and Remote Sensing, vol. 8(3), pp. 542-546, 2011.
 S. Serpico et G. Moser, «Extraction of spectral channels from hyperspectral images for classification purposes,» chez IEEE Transactions on Geoscience and Remote Sensing, 2007.
 A. Martinez Rios, N. Richard et C. Fernandez, «Alternative to colour feature classification using colour contrast occurrence matrix,» chez 12th Quality Control by Active Vision (QCAV), Dijon, 2015.
 R. Coliban, M. Ivanovici et I. Szekeli, «Color and multispectral texture characterization using pseudo-morphological tools,» chez ICIP, Paris, 2014.
 A. Ledoux, N. Richard, A. Capelle-Laizé, H. Deborah et C. Fernandez-Maloigne, «Toward a full-band texture features for spectral images,» chez ICIP, Paris, 2014.
 H. Deborah, N. Richard et J. Hardeberg, «A Comprehensive Evaluation of Spectral Distance Functions and Metrics for Hyperspectral Image Processing,» IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS), vol. PP(99), pp. 1-11, 2015.
 A. Ledoux, «Vers des traitements morphologiques couleur et spectraux valides au sens perceptuel et physique: méthodes et critères de sélection,» Université de Poitiers, Poitiers, 2013.
 N. Richard, D. Helbert, C. Olivier et M. Tamisier, «Pseudo-Divergence and Bidimensional Histogram of Spectral Differences for Hyperspectral Image Processing,» Journal of Imaging Science and Technology, vol. (to appear), pp. 1-13, 2016.
 H. Deborah, «Towards spectral mathematical morphology,» NTNU-Université de Poitiers, 2016.
 IND52, Multidimensional reflectometry for industry (EMRP Project xC-reflect).
 N. David, N. Pentinen, R. Urbas et M. Klanjsek Gunde, «Measuring gonioapparent samples using a bidirectional spectrometer and SRGB visualisation,» Prague, 2016.
(c) GdR 720 ISIS - CNRS - 2011-2015.