Ariana is a joint project-team of INRIA, and of CNRS and the University of Nice-Sophia Antipolis via the Computer Science, Signals and Systems Laboratory (I3S) in Sophia Antipolis (UMR 6070). It was created in 1998.

The Ariana project-team aims to provide image processing tools to aid in the solution of inverse problems arising in a wide range of concrete applications, mainly in Earth observation and cartography, for example cartographic updating, land management, and agriculture, while at the same time advancing the state of the art in the image processing methods used to construct those tools. Certain applications in biological and medical imaging are also considered, using the same tools as in remote sensing.

Extention of 2 patents in collaboration with Galderma.

Two prizes for Ariana students.

Ian Jermyn (CR1) left INRIA at the end of August 2010 to take up a position as Reader in Statistics at the University of Durham in the UK.

Following a Bayesian methodology as far as possible, probabilistic models are used within the Ariana project-team, as elsewhere, for two purposes: to describe the class of images to be expected from any given scene, and to describe prior knowledge about the scene in the absence of the current data. The models used fall into the following three classes.

Markov random fields were introduced to image processing in the Eighties, and were quickly applied to the full range of inverse problems in computer vision. They owe their popularity to their flexible and intuitive nature, which makes them an ideal modelling tool, and to the existence of standard and easy-to-implement algorithms for their solution. In the Ariana project-team, attention is focused on their use in image modelling, in particular of textures; on the development of improved prior models for segmentation; and on the lightening of the heavy computational load traditionally associated with these techniques, in particular via the study of varieties of hierarchical random fields.

The development of wavelets as an alternative to the pixel and Fourier bases has had a big impact on image processing due to their spatial and frequency localization, and the sparse nature of many types of image data when expressed in these bases. In particular, wavelet bases have opened up many possibilities for probabilistic modelling due to the existence of not one but two natural correlation structures, intra- and inter-scale, leading to adaptive wavelet packet models and tree models respectively. In Ariana, attention is focused on the use of tree models for denoising and deconvolution; adaptive wavelet packet models for texture description; and on the use of complex wavelets for their improved translation invariance and directional selectivity.

One of the grand challenges of computer vision and image processing is the expression and use of prior geometric information. For satellite and aerial imagery, this problem has become increasingly important as the increasing resolution of the data results in the necessity to model geometric structures hitherto invisible. One of the most promising approaches to the inclusion of this type of information is stochastic geometry, which is a new and important line of research in the Ariana project-team. Instead of defining probabilities for different types of image, probabilities are defined for configurations of an undeterminated number of interacting, parameterized objects located in the image. Such probability distribution are called `marked point processes'. For instance, two examples that have been developed in Ariana use interacting cuboids of varying length, width, height and orientation for modelling buildings; and interacting line segments of varying length and orientation for modelling road and other networks.

The use of variational models for the regularization of inverse problems in image processing is long-established. Attention in Ariana is focused on the theoretical study of these models and their associated algorithms, and in particular on the -convergence of sequences of functionals and on projection algorithms. Recent research concerns the definition of and computation in a function space containing oscillatory patterns, a sort of dual space to BV space, which captures the geometry of the image. These variational methods are applied to a variety of problems, for example image decomposition.

In addition to the regularization of inverse problems, variational methods are much used in the modelling of boundaries in images using contours. In Ariana, attention is focused on the use of such models for image segmentation, in particular texture segmentation; on the theoretical study of the models and their associated algorithms, in particular level set methods; and on the incorporation of prior geometric information concerning the regions sought using higher-order active contour energies.

Wavelets are important to variational approaches in two ways. They enter theoretically, through the study of Besov spaces, and they enter practically, in models of texture for segmentation, and in the denoising of the oscillatory parts of images.

One of the most important problems studied in the Ariana project-team is how to estimate the parameters that appear in the models. For probabilistic models, the problem is easily framed, but is not necessarily easy to solve, particularly in the case when it is necessary to extract simultaneously from the data both the information of interest and the parameters. For variational models, there are few methods available, and the problem is consequently more difficult.

These are perhaps the most basic of the applications with which Ariana is concerned, and two of the most studied problems in image processing. Yet progress can still be made in these problems by improving the prior image models used, for example, by using hidden Markov trees of complex wavelets or by decomposing the image into several components. Ariana is also interested in blind deconvolution.

Many applications call for the image domain to be split into pieces, each piece corresponding to some entity in the scene, for example, forest or urban area, and in many cases for these pieces to be assigned the appropriate label. These problems too are long-studied, but there is much progress to be made, in particular in the use of prior geometric information.

As the resolution of remote sensing imagery increases, so the full complexity of the scene comes to the fore. What was once a texture is now revealed to be, for example, an arrangement of individual houses, a road network, or a number of separate trees. Many new applications are created by the availability of this data, but efficient harvesting of the information requires new techniques.

Earth observation and cartography is not solely concerned with 2D images. One important problem is the construction of 3D Digital Elevation Models (DEMs) from high-resolution stereo images produced by satellites or aerial surveys. Synthetic Aperture Radar (SAR) imagery also carries elevation information, and allows the production of more accurate DEMs thanks to interferometry techniques, for example.

Every day, vast quantities of data are accumulated in remote sensing data repositories, and intelligent access to this data is becoming increasingly problematic. Recently, the problem of retrieval from large unstructured remote sensing image databases has begun to be studied within the project.

The software SARDecoder V1.1 was transfered to French Space Agency (CNES).

The software PHASEFLOW V1.0 was transferred to French Space Agency (CNES).

The software Blinde V2.0 was transferred to German Space Agency (DLR).

The software GRENAT V2.0 was transferred to CIRAD in Montpellier.

Two patents (OA09437 and OA09438) were extended jointly with GALDERMA at the end of 2010.

The software ASOE V1.0 was deposited with the APP in November 2010. It was developped for extracting multiple arbitrary shaped objects using stochastic multiple birth-and-death dynamics and active contours.

The software ThinBlinDe V1.0 was deposited with the APP in August 2010. It deals with blind deconvolution for confocal laser scanning microscopy for thin specimens.

The software PHASEFLOW V1.0 was deposited with the APP in March 2010. It deals with the extraction of directed networks from remote sensing images, based on a nonlocal phase/vector field model. The number of the parameters is estimated automatically.

The software SARDecoder V1.1 was deposited with the APP in January 2010. It was developed for classifying high resolution Single-Pol and Dual-Pol Synthetic Aperture Radar (SAR) images, in collaboration with G. Moser and S. Serpico from the University of Genoa in Italy, and V. Krylov from Moscow State University in Russia.

This work was performed in collaboration with Professor Zoltan Kato of the University of Szeged, Hungary [ http://www.u-szeged.hu/english/].

The phase field higher-order active contour framework
for shape modelling developed in the EPI Ariana lends
itself to a probabilistic interpretation, the phase field
energies being taken as the Gibbs energies of a Markov
Random Field (MRF). This opens the way to parameter and
model estimation, stochastic algorithms, and much
else
. However, one significant
limitation of the original framework remains: the inability
to represent overlapping objects. The representation used
in the phase field framework and its MRF equivalent, is of
a
*region*, not a set of objects. In this work we
overcome this limitation in the case of near-circular
objects, but it is clear that the mechanism used extends to
any model in the phase field/MRF shape modelling
framework.

We take the binary MRF model from
and extend it by adding
`layers',
*i.e.*the MRF becomes a map from the image domain to
{±1}
^{n}, where
n>1. The number
nis the size of a maximal cluster of mutually
overlapping objects. Thus each layer has an associated
binary field that specifies a region corresponding to
objects, while overlapping objects are represented by
regions in different layers. The model assigns high
probability to object configurations in the image domain
consisting of an unknown number of possibly touching or
overlapping near-circular objects of approximately a given
size. This is achieved by keeping from
the long-range interactions
favouring connected components of approximately circular
shape within each layer, while regions in different layers
that overlap are penalized by overlap area. If two nearby
objects exist on different layers, the mutual repulsion
produced by the long-range interactions is eliminated and
replaced by the short range overlap penalty.

Used as a prior coupled with a suitable data likelihood,
the model can be used for object extraction from images,
*e.g.*cells in biological images or densely-packed
tree crowns in remote sensing images. The first row of
figure
show ground states for various
numbers of layers
and overlap penalty
. As the number of layers increases, more and more
objects can be packed into the image domain, because the
mutual repulsion between neighbouring objects can be
eliminated if they live on different layers. As the overlap
penalty increases, the objects overlap less in the ground
state. The next row shows segmentation results obtained on
several synthetic images of circular `objects' with
different degrees of overlap. Note how even extremely
overlapping objects can be segmented as distinct. The
bottom row shows results obtained on cell imagery by the
University of Szeged.

This study was supported by INRIA Associated Team ODESSA [ http://www-sop.inria.fr/ariana/Projets/Odessa/index.html]. It was conducted in collaboration with Serguei Komech, IITP in Moscow [ http://www.iitp.ru].

In this work, we address shape classification. A shape
is a convex bounded set in
. We consider a basic descriptor
_{0}(
S)defined as the ratio of the
volume of the
-neighbourhood of the shape to the shape volume. The
initial shape is then transformed by a map, parameterized
by an angle
, which extends the shape along the direction
by a factor
and contracts the shape along the orthogonal
direction by a factor
. We thus obtain a function for our descriptor
(
S,
). We have defined a metric on
this descriptor space. We have shown that this metric is
continuous with respect to the Haussdorf metric in the
initial shape space. We have tested this metric for shape
retrieval on the MPEG-7 database (see figure
and table
) and on the Kima database. The
results are convincing for discriminating complex
shapes.

1st | 2nd | 3rd | 4th | 5th | 6th | 7th | 8th | 9th | 10th | |

Bonefull | 95 | 70 | 85 | 85 | 90 | 85 | 85 | 85 | 70 | 75 |

Heart | 100 | 100 | 100 | 100 | 100 | 100 | 95 | 95 | 95 | 100 |

Glas | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 90 | 100 | 95 |

Fountain | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 |

Key | 100 | 100 | 95 | 100 | 95 | 95 | 90 | 90 | 95 | 95 |

Fork | 95 | 90 | 65 | 70 | 65 | 75 | 65 | 75 | 70 | 60 |

Hammer | 95 | 95 | 80 | 30 | 30 | 35 | 30 | 40 | 35 | 15 |

This work has been made in collaboration with Pierre Couteron and Christophe Proisy from IRD Montpellier [ http://www.mpl.ird.fr/], and was partially supported by AAP INRA/INRIA [ http://www.inra.fr/].

The use of remote sensing information for studying large areas of heterogeneous forest arises new interest because of the increasing availability of data at very high spatial resolution (VHR), including optical data and LiDAR altimeter data of the canopy. On the other hand, the theoretical reflection on allometries and functions of scales involving the biomass, which attracts significant interest in integrative biology, has not previously benefited from the contribution of remote sensing and has faced only a very limited field data measurements in terms of trees (especially concerning the height and the sizes of crowns) and spatial structures of populations. This work lies at the interface between ecological modeling and pattern recognition in remote sensing images at very high spatial resolution (metric). The objective is to use and adapt the pattern recognition and identification of objects, including algorithms from stochastic geometry, to acquire a sufficiently broad and precise view of endogenous structuring of tropical forests through the visible structures in canopy (crown types, gaps, etcâ¦) on spatial imagery. This knowledge will be an essential contribution to the development of analytical models on both algometric properties of individual trees as the emergence of complex structures through interactions between trees. This is an interdisciplinary work combining image analysis and modelling form of plants. The main approach concerning image analysis is based on marked point processes, previously developed by EPI Ariana, and the alternative approaches are considered by global characterization of texture developed by UMR AMAP [ http://amap.cirad.fr/en/presentation.php]. An example of detection of trees with their canopy size through an optical image of mangrove forest is shown in figure .

This work has been made in collaboration with Pierre Couteron and Christophe Proisy from IRD Montpellier [ http://www.mpl.ird.fr/], and was partially supported by AAP INRA/INRIA [ http://www.inra.fr/].

This work is being performed in collaboration with Dr.
Michel Gauthier-Clerc from La Tour du Valat [
http://www.tourduvalat.org/] and Prof. Yvon Le Maho
from Institut Hubert Curien [
http://

This work is divided into two parts. The first part
focused on 3D object extraction from hand held camera
applied to the counting of penguins in a colony. We
proposed an object based method founded on the Marked Point
Process (MPP) theory because it easily allows modeling
scenes made of geometrically specified objects. The MPP is
based on the definition of a configuration space composed
of object sets, to which is attached a Gibbs energy
function. Minimizing this energy function leads to the
detection of our objects of interest. We use the recently
developed Multiple Birth and Death algorithm for the energy
minimization due to its speed advantage. The main
contribution of this work is a novel treatment of
occlusions. We introduce a new parent-child dependency
which extends the existing MPP using a mixed graph, this
extra dependency is due to occlusions. Most of previous
works
*e.g.*on pedestrians, object (human) geometry is
approximated either using a rectangle or an ellipse. We
propose to use a fine quality 3D geometrical model for
objects (penguins), and the usage of OpenGL for projection,
taking advantage of the graphics card GPU. We validated our
model on semi-synthetic images.

The second part of this work consists in
the introduction of a new optimization method for MPP, that
we call Multiple Birth and Cut (MBC)
. This method is based on the
multiple birth and death algorithm and the popular graph
cut algorithm. The birth part consists of proposing
configuration of non-overlapping
*e.g.*ellipses, for which a data term and prior term
is calculated. From the proposed configuration, in the
death step of the algorithm, we keep good object and remove
non fitting ones, based on the cut of a special graph. We
validated our model on the flamingo counting inside
colonies, showing that our method overcomes the detection
obtained by the MBD algorithm
. Moreover, we avoid the non
trivial task of tuning the simulated annealing
parameters.

This work is being performed in collaboration with Mathieu Brédif and Marc Pierrot-Deseilligny from Matis laboratory, IGN [ http://www.ign.fr].

The automatic detection and reconstruction of the urban vegetation is of major interest in urban planning. Climate, noise probation and other environmental models used in modern urban planning demand, as input data, information about the tree areas of cities. This work proposes a new fully automatic approach for simultaneous detection and extraction of building footprints and tree crowns in a dense urban environment. Elevation data provided from a photogrammetricaly derived Digital Elevation Model (DEM) and radiometric information from color infrared orthophotos are used as input. The buildings are extracted as rectangles and the tree crowns as circles with a Multi-Marked Point Process (MMPP) sampled by a Reversible Jump Monte Carlo Markov Chain sampler (RJMCMC) coupled with a simulated annealing process in order to optimize an energy function. The method allows the simultaneous and automatic modeling of complex environments without any prior knowledge about the number or the position of the different elements that constitute the urban environment. This approach allows the representation of various forms of building and trees in a simplified yet efficient way. Some primilinary results of the algorithm are presented on both a DEM and an NDVI image in Figure .

This study was supported by INRIA Associated Team ODESSA [ http://www-sop.inria.fr/ariana/Projets/Odessa/index.html] and an ECONET project. It was conducted in collaboration with P. Lukashevich, A. Krauchonak, and B. Zalesky from the UIIP in Minsk [ http://www.uiip.bas-net.by/index-eng.html], E. Zhizhina from the IITP in Moscow [ http://www.iitp.ru], and J.D. Durou from IRIT in Toulouse [ http://www.irit.fr/?lang=en].

In this project, we aim at reconstructing buildings in 3D from one or several aerial or high resolution satellite images. The main idea is to avoid solving the so-called inverse problem. We will simulate configurations of buildings and test them with respect to the data. The generation of configurations will be performed using multiple birth-and-death dynamics . A Gibbs point process is defined including prior information about building configurations. To define the data term, the building configuration is projected into the data plane(s), using models of shading and shadows. This projection is performed using OpenGL for a fast 2D rendering of the scene. The data term is based on the consistency of shadows in the image and on the configuration projection in the image plane, whereas the prior penalizes building overlaps. The preliminary results are encouraging. The next steps consist of refining the data model by embedding information about gradients, and improving the convergence speed by defining proper birth maps for generating new buildings.

This work is conducted in collaboration with Matis laboratory, IGN [ http://www.ign.fr].

This work consists in generating high resolution Digital Elevation Models (DEM) from Lidar point clouds by taking into account urban knowledge on the observed scenes. The generation of such semantized DEMs is lead by a point cloud classification which separates buildings facets, building edges, vegetation, ground and urban details such as cars, chimneys. This labeling problem is solved by using a Markov Random Field formulation. The optimal labeling is found by Graph-Cuts. Then, the obtained point cloud classification is used to create a dense and semantized Digital Elevation Model by locally adapting the regularization procedure with respect to the expected urban objects.

This work is done in collaboration with Matis laboratory, IGN [ http://www.ign.fr].

The generation of 3D representations of urban
environments
from aerial and satellite data
is a topic of growing interest in image processing and
computer vision. Such environments are helpful in many
fields including urban planning, wireless communications,
disaster recovery, navigation aids, and computer games.
Laser scans have become more popular than multiview
aerial/satellite images thanks to the accuracy of their
measurements and the decrease in the cost of their
acquisition. In particular, full-waveform topographic LIDAR
constitutes a new kind of laser technology providing
interesting information for urban scene analysis
. We study new stochastic
models for analysing urban areas from LIDAR data (see
Figure
). We aim to construct concrete
solutions to both urban object classification (
*i.e.*detecting buildings, vegetation, etc.) and the
3D reconstruction of these objects. Probabilistic tools are
well adapted to handling such urban objects, which may
differ significantly in terms of complexity, diversity, and
density within the same scene. In particular,
jump-diffusion based samplers offer interesting
perspectives for modelling complex interactions between the
various urban objects.

This work was partially funded by French Space Agency (CNES)[ http://www.cnes.fr/].

The problem of feature extraction from high resolution remote sensing images has been addressed in several fields using different approaches. One of the most successful approaches is marked point process modeling. In fact, in a marked point process framework, the objects of the image are represented by a set of interacting geometric shapes. This object set is governed by two types of energy: a data energy term which links the objects to the features to be extracted and a regularizing energy term which controls the repartition of objects in the scene. This model incorporates some parameters which allow us to model strong connections between objects according to the processed image. In order to achieve unsupervised object extraction, we need to develop an estimation method of those parameters.

Previously, an estimation method based on the Stochastic Expectation-Maximization algorithm was studied and proved its relevance for estimating these parameters. It was only validated on a simple model of a marked point process of circles.

We have first extended this estimation procedure to more general geometrical shapes such as ellipses and rectangles . Different types of objects have been successfully extracted namely flamingos, tree crowns and building footprints.

Then, we have dealt with the problem of boat counting using an ellipse model and the detection of refugee tents using a rectangle model. We have proposed new prior and data terms for boat detection. In fact, boats in a seaport are very close and aligned, which makes their discrimination difficult using the model proposed in . Moreover, we have modified the data energy component for tent detection since the considered model is based on the object geometry and does not take into account other type of information such as the object color , . Figure shows the result obtained for boat detection.

This work is done in collaboration with DIBE, University of Genoa with Gabriele Moser and Sebastiano Serpico [ http://spt.dibe.unige.it/] with partial support of the French Defense Agency, DGA [ http://www.defense.gouv.fr/dga/] and the Italian Space Agency, ASI [ http://www.asi.it/en].

Synthetic Aperture Radar (SAR) techniques improved, and it is now possible to acquire very high resolution (VHR) images. This high resolution (around 1 m) combined with a short revisit time (up to 12 hours) are precious information so as to monitor urban areas and infrastructures, especially critical with respect to natural disasters. The main problems of SAR imagery are the speckle noise and the shadows, that is why good optical images classification methods are not adapted to SAR ones, above all in urban areas. The developed method combines the Markov Random Field (MRF) approach to Bayesian supervised classification and the Dictionary-based Stochastic Expectation Maximization (DSEM) approach to SAR amplitude Probability Density Function (PDF) estimation.

The first step of the method consists in modeling the SAR amplitude statistics as a finite mixture of parametric components automatically drawn from a dictionary of SAR specific PDFs. When applied to each class-conditional PDF in a VHR image, DSEM represents a natural model for the related heterogeneity, leading to a mixture estimate where distinct components may be interpreted as the contributions of different ground materials (e.g., roofs, concrete, water, grass). The second step deals with the classification itself. In order to incorporate contextual information and gain robustness against speckle, a hidden MRF approach is considered. Spatial regularization parameters in Potts energy function are estimated by an accelerated simulated annealing algorithm. In order to generate the output classification map, the energy function is minimized by a modified Metropolis dynamics algorithm.

Texture features such as GLCM (Grey Level Co-occurrence Matrix) textures, extracted from the original SAR image, turn out to discriminate quite well the urban areas (see figure ). To combine both SAR amplitude and textural feature data, a copula-theoretic approach is used to estimate their joint statistics. Specifically, the flexibility of DSEM, granted by its essentially nonparametric formulation, makes it feasible to estimate the marginal PDFs of both the amplitude and the texture feature. Copula functions allow a joint bivariate PDF to be modeled, given the related marginal PDFs. The resulting joint PDF estimates are plugged into the MRF model considered above.

The method was tested on real COSMO-SkyMed images . We illustrate the obtained results with an example of a SAR acquisition in the region of Cavallermaggiore (Italy). Spatially disjoint training and test areas were manually annotated. The classification is done following 3 classes: urban areas, natural landscape and wet areas. The results are shown qualitatively in figure . The computation of numerical results gives an average accuracy of 98.9 percent for the considered test areas.

This work is conducted in collaboration with DIBE, University of Genoa with Gabriele Moser and Sebastiano Serpico [ http://spt.dibe.unige.it/] with the support of the French Space Agency, CNES [ http://www.cnes.fr], the Italian Space Agency, ASI [ http://www.asi.it/en], the Poncelet laboratory in Moscow [ http://www.mccme.ru/lifr/] and INRIA-DRI.

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) image
processing. The research here focused on the supervised SAR
image classification, which is one of the fundamental SAR
image processing problems. Recently, various models have
been proposed for modeling the single channel statistics of
SAR data, however, none of them general and flexible enough
to model the joint probability density function (PDF) in
case of
D-channel SAR,
. We propose a joint PDF model for multichannel SAR,
based on finite mixture modeling for marginal PDFs
estimation and copulas for multivariate distribution
modeling
. We apply this model to medium
and high resolution multichannel SAR amplitude image
classification by combining it with a contextual Markov
Random Field approach (MRF) that allows to take into
account the contextual information and to gain robustness
against the inherent noise-like phenomenon of SAR known as
speckle. The finite mixture modeling is done via a recently
proposed SAR-specific Dictionary-based Stochastic
Expectation Maximization (DSEM) approach, that is applied
to class-conditional amplitude probability density function
estimation 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. The
contribution of this study is the generalization of the
recently considered DSEM - MRF classification approach to
D-channel SAR,
, via copulas.

The developed Copula-DSEM-MRF approach has been
experimentally validated on several multichannel Quad-pol
RADARSAT-2 images and compared with a benchmark “
K-nearest neighbors” (
K-NN) classification technique, combined with MRFs (
K-NN-MRF), see, e.g., Fig.
. These experiments demonstrate
the developed model to be flexible and perform well on
multichannel SAR and to significantly outperform the
benchmark approach on urban class.

This work is done in collaboration with Prof. Anuj Srivastava, Dept. of statistics, Florida State University [ http://www.fsu.edu/] and is funded by the EADS Foundation [ http://www.fondation.eads.net/en/].

Shape modelling is a subject of great importance in image processing and computer vision: there are many segmentation problems for which prior knowledge of object shape is essential to solving the problem. Higher-order active contours (HOACs) incorporate sophisticated shape information via long-range interactions between points of the region boundary. This allows a single model to describe a range of objects, eliminates pose estimation, and permits the detection of multiple object instances, while constraining object shape.

So far, the HOAC framework has been used to model simple shapes, e.g. networks for road segmentation and circles for tree crown segmentation . This work consists in developing a way to model more complex shapes within this framework, without losing its advantages. To model a particular shape, a stability analysis of the shape is performed. Perturbations of the shape are expressed in terms of Fourier components on its boundary, and the energy defining the HOAC model is expanded to second order in these perturbations. If parameter values exist such that every perturbation increases the second-order energy, then the shape is a local minimum and hence stable.

While modelling simple shapes, we have observed in numerical experiments the formation of other, more complex stable shapes. The first step is to verify that stable non-zero amplitudes exist via higher-order stability analysis of simple shapes. If they do exist, then HOACs can be used to model complex shapes by adjusting the long-range interactions to give preferred amplitudes to each Fourier component (see figure ). The next step is to control these amplitudes by inverting the energy expansion to give the interaction function in terms of the amplitudes. These theoretical analysis will be compared to numerical experiments to test their consistency.

This work is funded by a PACA regional grant and by INRIA via a contract with the French Space Agency [ http://www.cnes.fr], in collaboration with Thales Alenia Space [ http://www.thalesgroup.com].

This work concerns a new family of phase field models
designed to model the shape of `directed networks' in
images (
*e.g.*vascular networks in medical imagery and
hydrographic networks in remote sensing imagery), and to be
used for the segmentation of such networks
. The model extends existing
phase field models of undirected networks (
*e.g.*road networks)
,
, via the inclusion of a vector
field representing the `flow' running through the network,
in addition to the usual scalar phase field describing the
network via its smoothed characteristic function. The
presence of the vector field allows the model to
incorporate characteristic geometric properties of directed
networks.

The directed network model has a large number of free, unphysical parameters, which makes parameter learning very difficult. In particular, the model can favour geometric configurations other than networks for significant parameter ranges. In this work, we analysed the stability of a directed (because the vector field is present) long, straight bar to perturbations of its boundary. The analysis yields constraints necessary for stability of the bar: the constraints eliminate some parameters, replace others by physical parameters such as bar width, and place bounds on the remainder. The parameters can thus be tuned to values that favour linear structures for network modelling. Figure shows the results of the stability analysis leading to the parameter constraints, as well as segmentations of hydrographic networks from very high resolution images , showing that the directed network model outperforms the undirected network model.

This work is funded by the French Space Agency, CNES [ http://www.cnes.fr] and Thales Alenia Space [ http://www.thalesgroup.com] in collaboration with Marc Antonini from I3S/UNS [ http://www.i3s.unice.fr/I3S/presentation.en.html].

Current acquisition chain consists in sampling the image (giving N coefficients) and in compressing it to retain only K coefficients (with K << N). We have first focussed our work on the acquisition system and are currently investigating an acquisition chain which would directly give the K coefficients with the same idea as the Compressed Sensing (CS) method. Current acquisition system uses the sparsity of the Discrete Wavelet Transform (DWT), to compress the image. Compressed Sensing considers on the contrary transforms which give flat spectrum and use the property of sparsity to recover the image exactly with less coefficients than Shannon. Following the idea of saving only these M coefficients, we are also investigating a wavelet transform which would give only the M higher coefficients (DWT method).

We are considering such transforms and use noiselets transform in the case of the CS method and a 3 level 9/7 wavelet transform for the DWT method. Results are shown in figure . The Peak Signal To Noise Ratio is used to compute the distortion between the retrieved and the original image while the rate specifies the number of bits needed to code the image. To get an upper bound of the efficiency of the methods, we consider the image to be sparse in the singular value decomposition which represents an ideal case as it supposes the knowledge of the decomposition matrices. The results are compared with the current method recommended by the Consultative Committee for Space Data Systems (CCSDS) which consists in a discrete wavelet transform, where all the coefficients are retained, followed by a coding step and an inverse transform. From a general point of view, it appears first that coefficients of Compressed Sensing method are not well suited to coding (too large distribution). However we get convincing results with this method, as shown in figure , if we are able to get a sparse representation of the image in some basis. Globally, we can see that both methods outperform the CCSDS method by taking less measurements than needed, the DWT method giving better results for many measurements. However, we get an interesting result if we can select only few coefficients. Indeed, with the CS method, only 10% of measurements of the image encoded at 3 bits/pixel allow to reconstruct an image with a PSNR of 60 dB. The same value of PSNR is reached by the CCSDS method only at 6 bits/pixel. These methods may thus be efficient for the limited resources of on-board satellite hardware.

This work is funded by the ANR Define Diamond in collaboration with Alain Dieterlen from UHA, Mulhouse [ http://www.uha.fr/], Jean-Christophe Olivo-Marin and Praveen Pankajakshan from Pasteur Institute, Paris [ http://www.pasteur.fr/ip/easysite/go/03b-00002j-000/en], Gilbert Engler from INRA, Sophia Antipolis, and Prof. Zvikam, Weizmann Institute, Rehovot, Israel.

During the last decade, biological imagery systems got a considerable advance providing biological specimen images at tissue and cellular level. One of the most common systems is the confocal laser scanning microscopy which allows a 3D visualization of living specimen at resolutions of about one hundred of nanometers. Although it offers higher resolution and better contrast compared to the wide field microscopy, it suffers from some artifacts. In fact, even under the most suitable imaging conditions, specimen images are affected by blur mainly due to the inherent diffraction limited of the optical system, spherical aberration, and the incoming out of focus light. Hence, such a problem can be viewed as a convolution of the real object and the Point Spread Function (PSF) of the optical acquisition system. In order to restore the original image, a previous study of blind deconvolution methods was carried out in a Bayesian framework . A parametric model of the PSF was established under some approximations. The estimation of both the PSF parameters and the object is achieved by an Alternate Minimization (AM) scheme.

In this work, we develop a new restoration method of the original fluorescence image using another imaging technique. We use images obtained by an optical diffractive tomographic microscopy to regularize the estimated object. It offers indeed complementary details from the observed specimen. In fact, diffractive tomographic microscopy is a 3D imaging technique that allows to measure the refraction index distribution within the specimen. We thus look for models of image formation since biological specimens have heterogeneous refraction index distribution. Some aberrations are induced, observed images are therefore deformed.

This work is performed in collaboration with Sebastien Schaub, head of the microscopy platform at IBDC [ http://www.unice.fr/ibdc/].

The purpose of this work is to develop numerical restoration methods for 3D images from confocal microscopy. Even though confocal microscopy removes a lot of blur present in widefield microscopy images, the confocal images still show blurring in the depth direction in addition to Poisson noise due to the small number of photons. Restoration methods adapted to the Poisson statistics of these images have been developed in the past in the EPI Ariana. These methods are based on the Richardson-Lucy algorithm to which we add a priori information such as total variation.

In this work, we generalize Richardson-Lucy with total variation to take into account an additional observed image. It corresponds to specific acquisitions where we have two images, with two different noise realization of the same specimen. This leads to the minimization of a functional formed by two data fidelity terms and one regularization term.

The weights between the two data fidelity terms and the regularization term are difficult to set. Then we propose to minimize the regularization term under constraints on the observations given by the noise. We first developed a Gaussian approximation of the noise and solved the constrained minimization problem by minimizing the augmented Lagrangian in alternate directions after variable separation. The same idea is currently being studied in the case of Poisson noise.

Results of this generalized algorithm are shown Figure using images of Figure .

This work was partially funded by a contract with Galderma [ http://www.galderma.com].

Spectral imaging of the skin is used to quantified as precisely and as quickly as possible the degree of a disease to evaluate the efficiency of a treatment. In this work, the studied disease is Melasma. It is a localized skin hyperpigmentation appearing on the face. For skin analysis, several methods had been proposed in the literature. Most of them, like the Stamatas algorithm, are based on models of chromophores absorbency depending on the wavelength.

Our approach is to use signal processing methods to quantify the hyperpigmentation of the skin , . Then, the general treatment chain can be divided in three parts.

The first one, the shading compensation, is used to compensate the artefacts introduced by the volume of the pictured face. An empirical model and a model based on interferences and waves reflections on Lambertian areas have been studied.

The second step, the data reduction by projection pursuit, allows to reduce the quantity of spectral information by filtering the redundancy and therefore avoiding the Hughes phenomenon.

The third step, the classification, is done with a supervised Support Vector Machine (SVM). This classification scheme was compared to an ICA based method and the Stamatas algorithm in .

In order to automate the classification, we introduce a spectral analysis step based on an index or distance. This index browses the spectrum to measure it variations. Then outliers of the spectral variation distribution are selected to train the SVM. Figure shows the quantification map of hyperpigmentation obtained with the proposed methods based on SVM (AS-PP-SVM), an ICA based technique and the Stamatas algorithm. For the SVM based method, we assume that the distance between the SVM separating hyperplane and the pixel value is proportional to the hyperpigmentation degree. AS-PP-SVM provides as accurate disease quantification as the Stamatas algorithm and has the advantage of providing a threshold between the healthy and the hyperpigmented skin.

Contribution of multi and hyperspectral imaging to skin pigmentation evaluation. Contract #4383.

Detection of objects in infrared imagery using phase field higher-order active contours. Contract # 4643.

Development of advanced image-processing and analysis methods as a support to multi-risk monitoring of infrastructures and urban areas. Grant from the French Defense Agency, DGA.

Parameter estimation of marked point processes for object extraction from high resolution satellite images. Contract #2150 part 1.

Higher-order active contours with application to the extraction of networks (roads and rivers) from high resolution satellite images. Contract # 2150, part 2.

Modelling of high resolution SAR image statistics. Contract # 4635

Optimization of the compression-restoration chain for satellite images. Grant from CNES and TAS.

Global reconstruction of urban areas. Grant from IGN.

Airbone devices for survey and detection. In collaboration with ATE (PI), Dronexplorer, Nexvision, Coreti. This project has been labelled by the `pôle Pegase'.

In collaboration with Thales Alenia Space (Toulouse).

In collaboration with the J. A. Dieudonné Laboratory of CNRS/UNS (Gilles Aubert, Luis Almeida, David Chiron, Laurence Guillot), the Pasteur Institute (Jean-Christophe Olivo-Marin), and SAGEM DS Argenteuil (Yann Le Guilloux, Daniel Duclos).

In collaboration with IMFT (F. Plouraboue (PI), R. Guibert), CERCO (C. Fonta), and ESRF (P. Cloetens, G. LeDuc, R. Serduc).

In collaboration with the Pasteur Institute (Jean-Christophe Olivo-Marin, Praveen Pankajakshan), the MIPS laboratory of Université de Haute Alsace (Alain Dieterlen, Bruno Colicchio), the LIGM of Université Paris-Est (Caroline Chaux, Jean-Chritophe Pesquet, Hugues Talbot), and INRA Sophia-Antipolis (Gilbert Engler). This project has been labelled by the “pôle Optitec” and “pôle BioValley”. Web site: http://www-syscom.univ-mlv.fr/ANRDIAMOND.

In collaboration with Clément Mallet and Bruno Vallet from MATIS Laboratory, IGN [ http://www.ign.fr].

In collaboration with Pierre Couteron[PI], Christophe Proisy and Nicolas Barbier from UMR AMAP, IRD, INRA [ http://www.mpl.ird.fr].

In collaboration with G. Moser and S.Serpico[PI], from the University of Genoa (DIBE) and the Italian Space Agency (ISA).

In collaboration with the Dobrushin Laboratory of the Institute for Information and Transmission Problems of the Russian Academy of Science, Moscow (E. Zhizhina (PI), E. Pechersky, R. Minlos, S. Komech), the Image Processing and Pattern Recognition Laboratory of the United Institute of Informatics Problems of the National Academy of Science of Belarus, Minsk (B. Zalesky (PI), P. Lukaskevich, A. Krauchonak), and IRIT, Toulouse (J.D. Durou). Web site: http://www-sop.inria.fr/ariana/Projets/Odessa.

In collaboration with A.Srivastava, Department of Statistics, Florida State University, USA.

The members of the Ariana project-team participated actively in GdR ISIS and GdR MSPCV.

The Ariana project-team organized numerous seminars in image processing during 2010. 13 researchers were invited from the following countries: Austria, France, Iceland, India, Hungary, Portugal, Romania, Switzerland. For more information, see the Ariana project-team web site.

Members of the Ariana project-team participated actively in the visits to INRIA Sophia Antipolis of students from the Grandes Écoles (ENPC, ISAE/SUPAERO, ENS Cachan, KTH Stockolm), helped students of the Classes Préparatoires with TIPE in France; and gave information on remote sensing image processing to high school students in Mauritius.

Saima Ben Hadj attended a spring school entitled “Inverse problems in signal and image processing” in Porquerolles, France. She gave a talk at Sup'Com Tunis students in June and presented a paper at the conference PCV'10, Paris, in September.

Laure Blanc-Féraud gave a lesson at the summer school on `Inverse Problems in Signal and Image Processing' in Porquerolles, France, and attended summer school on "Apprentissage pour le TSI" in Peyresq. She attended the IEEE ICASSP'10 where she presented a poster and chaired a session. She gave an invited talk at the colloque STATIM on "Restauration de problèmes inverses et estimation d'hyperparamètres". She presented a poster at "Grand colloque STIC" of ANR in january, and organized the end meeting of ANR DETECFINE. She organized regularly meetings for Gyrovision FUI with ATE, and attended meetings with I3S, CNES and TAS for M. Carlavan PhD thesis supervision. She attended meetings of ANR DIAMOND in Sophia Antipolis, Mulhouse and Paris.

Xavier Descombes visited the Dobrushin Laboratory (IITP, Russian Academy of Science) for two weeks in July. He was an invited speaker in the Workshop on `Stochastic approaches for image processing' at CIRM in Luminy in May and gave an invited talk at GdR Isis in May. He organised several meetings for the Associated Team ODESSA and participated in ANR Micro-Réseaux project meetings. He regularly took part in meetings with Galderma.

Aymen El Ghoul gave seminars at the ADSTIC meeting at I3S, and at the SHAPE working group meeting, both in Sophia Antipolis, in April and May respectively. He presented a paper at the conference PCV/ISPRS'10 in Paris, France, in September.

Ahmed Gamal-Eldin presented the Ariana team in the “Euro-Mediterranean Innovation Marketplace 2010“ at Cairo in Egypt, where France was guest of honor. In May, Ahmed Gamal-Eldin has presented his masters work about fuzzy segmentation at ISSPA conference in Malaysia. He presented his PhD work at SITIS Conference again in Malaysia in December.

Ian Jermyn gave a seminar in the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, USA, in April. He gave invited talks in the Fifth Solar Image Processing Workshop, Les Diablerets, Switzerland and the International Workshop on Shape Perception in Human and Computer Vision, in conjunction with the European Conference on Computer Vision, Heraklion, Crete, Greece, both in September.

Vladimir Krylov gave a talk at the ORFEO Methodology meeting at CNES, Paris in January. He presented a paper at the conference 30th International Workshop on Bayesian Inference and Maximum Entropy Methods in Science and Engineering (MaxEnt), Chamonix, France in July.

Florent Lafarge gave seminars at the Institute for Computer Graphics and Vision , University of Technology , Graz, Austria, at the "journée GDR ISIS modélisation mathématique des textures", Paris, France and at the "journée envol recherche de la Fondation d'Entreprise EADS", Paris, France. He gave an oral presentation at CVPR'10, San Francisco, US.

Sylvain Prigent presented a paper at the conference Whispers'10 in Reykjavik (Iceland) in June, and presented a paper at the conference ICIP'10 in Hong Kong (China) in September.

Josiane Zerubia gave talks at CS,
Noveltis and Vega Technologies in Toulouse in January,
and participated in the CNES `Research and Technology
Day' in Labege and she also visited Telecom Paris-Tech,
in Paris. In February, she visited Astrium/EADS and
Sanofi Aventis, in Toulouse. In March, she visited again
Telecom-ParisTech, attended the annual Editorial Board
meeting of SFPT at CNES in Paris, and participated in the
IGN `Research Days' in Saint Mandé, and went to Tunisia
where she was part of two Masters Commitees at
ENSI-Tunis. She gave a plenary talk at PHARES 2010 [
http://

Laure Blanc-Féraud is Associate Editor of the “revue Traitement du Signal”. She also reviews papers for IEEE Trans. on Signal Processing and Applied Optics, Traitement du Signal and for the conferences IEEE ICIP, ICASSP, and ISBI as member of the IEEE BISP TC. She was a member of AERES visiting comitee of LIRIS laboratory, and member of a selection committee in Strasbourg University. She reviews proposals for the ANR Programme Blanc, and PEPS projects for CNRS. She was a reviewer for one HdR thesis and four PhD theses, and president of one PhD committee.

Xavier Descombes was a regular reviewer for IEEE TIP, IEEE TPAMI, IEEE TGARS, Traitement du Signal and IJRS. He was a reviewer for the CIBLE programme of Région Rhones Alpes. He was reviewer of two PhD thesis and one HdR.

Ian Jermyn was a reviewer for the journal IEEE TPAMI and for the conference ECCV'10.

F. Lafarge was a referee for Springer Eds., IEEE TPAMI, IEEE TIP, IEEE TSP, JPRS, PERS, IEEE CVPR, IEEE ICIP, ACIVS, GEOBIA. He reviewed proposals for the Czech Science Foundation.

Josiane Zerubia was president of one PhD committee, and a committee member for two more. She was a regular reviewer for IEEE TGRS, GRS Letters, and SFPT (Revue Française de Photogrammétrie et de Télédétection). She was a reviewer or a program committee member for ICASSP'10, ISBI'10 and ICIP'10, as member of the IEEE BISPTC and IEEE IVMSP TC, and for SPIE-ISPRS'10 (`Image and Signal Processing for Remote Sensing'), ISPRS PCV'10, GEOBIA'10, ICPR'10. Finally, she was part of the CR2 and CR1 selection comittee at INRIA Sophia-Antipolis Méditerrannée.

Laure Blanc-Féraud is a member of the IEEE Biological Image and Signal Processing Technical Committee. She is the `directrice adjointe' of the GdR ISIS. She is member of thematic consultation group of the Ministery of Research and Teaching (MESR) for "Math-Stic". She is a permanent member of the Organizing Committee of the Gretsi Peyresq annual Summer School in TSI. She is Associate Editor of the journal `Traitement du Signal' (Hermès). She is member of the steering committee of GdR "Mathématiques des Systèmes Perceptifs et Cognitifs" (MSPC). She is part of the CNECA 3 (equivalent of CNU for agricultural ministry). She is part of the Administrative Council of Gretsi. She is a member of the board of INRIA-SAM "Comité des projets".

Xavier Descombes is a member of the scientific committee of the `Pôle de compétitivité Optitec', and a member of the strategic committee of PopSud. He is computer systems coordinator for the Ariana project. He is PI of the Associated Team ODESSA.

Ian Jermyn was a member of the Programme Committee for the IEEE Computer Society Workshop on Perceptual Organization in Computer Vision. He is a member of the Doctoral Oversight Committee at INRIA Sophia Antipolis, and of the International Relations Working Group of the Scientific and Technological Orientation Council of INRIA. He is co-computer systems coordinator for the Ariana project.

Josiane Zerubia is an IEEE Fellow. She
is publicity chair of IEEE ICIP'11 in Brussels [
http://

Laure Blanc- Féraud is director of the "module de traitement numérique des images" of Master 2 at Poly'Tech Nice-Sophia Antipolis (UNS) and gave 17h. She also taught in the Biocomp Masters programme at Poly'Tech Nice-Sophia Antipolis (UNS) (12h) and in IMEA Master 1 in Valrose (20h).

Mikael Carlavan was teaching assistant for `Traitement Numérique du Signal' (18h) at Poly'Tech Nice-Sophia Antipolis.

Giovanni Gherdovich was teaching assistant for `Image/Compression Project' (20h) and `Introduction to Programming' (44h) at Poly'Tech Nice-Sophia Antipolis.

Xavier Descombes taught `Image analysis' (10h) at Poly'Tech Nice-Sophia, and `Image processing' and `Advanced techniques in space imagery' (20h) at ISAE/SUPAERO.

Ian Jermyn taught `Image analysis' (10h) at Poly'Tech Nice-Sophia, and `Filtering and segmentation of space imagery' (7.5h ETD) at ISAE/SUPAERO.

Florent Lafarge taught `Image analysis' (9h) at Poly'Tech Nice-Sophia, and `Image processing' and `Advanced techniques in space imagery' (7.5h) at ISAE/SUPAERO.

Josiane Zerubia was director of the module `Deconvolution and denoising in confocal microscopy' for the Masters 2 course BioComp at the University of Nice-Sophia Antipolis (24h of which 12 taught). She was director of the course `Advanced techniques for space imagery' at ISAE/SUPAERO (40h, of which 20h taught).

Jia ZHOU taught mathematics (60h) at IUT Montpellier.

Yannick Verdié: `Urban scene reconstruction from 3D point clouds', University of Nice-Sophia Antipolis. Defence expected in 2013.

Mikael Carlavan: `Optimization of the compression-restoration chain for satellite images', University of Nice-Sophia Antipolis. Defence expected in 2012.

Athanasis Georgantas: `Global reconstruction of urban scenes', EDITE, Telecom Paris-Tech. Defence expected in 2012.

Sylvain Prigent:`The contribution of multi and hyperspectral imaging to skin pigmentation evaluation', University of Nice-Sophia Antipolis. Defence expected in 2012.

Aurélie Voisin: `Development of advanced image-processing and analysis methods as a support to multi-risk monitoring of infrastructures and urban areas', University of Nice-Sophia Antipolis. Defence expected in 2012.

Jia Zhou: 'The contribution of object recognition from forest canopy images to the construction of an allometric theory of the structure of trees and of natural, heterogeneous forests', University of Montpellier 2. Defence expected in 2012.

Ahmed Gamal-Eldin: `Marked point processes models of 3D objects: an application to the counting of King Penguins, University of Nice-Sophia Antipolis. Defence expected in 2011.

Aymen El Ghoul: `Phase fields for the extraction of networks from remote sensing images', University of Nice-Sophia Antipolis. Defence expected in September 2010.

Ahmed Gamal Eldin received in December
the Best Paper Award at SITIS'10 Conference in Kuala
Lumpur, Maleysia. (
http://

Gregoire Kerr received in November the prize of the city of Toulouse for the best ISAE-SUPAERO Master Internship for his research work conducted at INRIA (EPI Ariana) in collaboration with Florida State University (Professor A. Srivastava).