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

The Ariana project 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. Another application concerns biological imaging using the same tools as in remote sensing.

The associated team SHAPES was created on january 1st 2007 (
http://

Josiane Zerubia has been nominated “Associate Editor-European Space Organizations and Industries, Space Signal and Image Processing” of the electronic journal Earthzine (
http://

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 field.

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 indeterminate 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.

Software for tree crown extraction from orchards using a marked point process model based on circles. It finds the positions of the trees. Deposited with the APP. This software has been transferred to the Joint Research Centre of the European Commission in Ispra, Italy.

Software for detecting flamingos in order to count them, using a marked point process model based on ellipses. Deposited with the APP. This is free software distributed under the CECILL C license. It has already been transferred to two ecological centres (Tour du Valat in Camargue, France and in Tunisia) and to the French Space Agency (CNES).

Software to extract circular tree crowns from panchromatic or colour infrared aerial images. The software is based on the `gas of circles' phase field model. Deposited with the APP and transferred to the Joint Research Centre (JRC) of the European Commission in Ispra, Italy and to the Hungarian Central Agricultural Office, Forestry Administration (CAO, FA) in Budapest, Hungary.

This software was deposited with the APP in 1999 (V1.0 with Anne Lorette, Ph.D. student, as co-author) and in 2003 (V2.0 with Osca Viveros-Cancino, Ph.D. student, as co-author). It has previously been transferred to the French Space Agency (CNES) and to Alcatel, and was transferred in 2007 to IRD, Earth Observation Division, in Montpellier.

This work is funded by the French National Forest Inventory (IFN) [
http://

Forests have been the subject of many studies because of their key role in the Earth's life cycle, particularly in the equilibrium of ecosystems and the balance of CO in the biosphere. Consequently, it is important to have good tools for monitoring the evolution of forest resources in the current climatic and industrial context. An increasing number of the methods being developed are based on remote sensing techniques. Forestry science uses remote sensing high resolution images for photo-interpretation, in order to establish inventories and maps, comparing this information with that acquired in the field. The work described here proposes different methods, based on a probabilistic approach, for extracting information from Colour InfraRed (CIR) aerial images. Two models for tree crown extraction, based on object processes, have already been studied in Ariana (see the 2005 and 2006 activity reports). In dense areas, an ellipse process is used, while in sparse zones, an ellipsoid process gives more information. As a result, we can obtain the number of trees, their position, their height, and the diameters of the crowns for sparse areas. The models have been tested on high resolution CIR aerial images provided by the French National Forest Inventory (IFN). The software GRENAT, developed in Ariana, implements the algorithms of these models, and many other specific tools, with an easy-to-use interface. This year, we worked on a user interface defined in collaboration with the French National Forest Inventory (IFN) for use by their operators.

This study was partially supported by CNRS and INRIA and has been conducted in collaboration with Prof. E. Zhizhina, IIPT Moscow (Russian Academy of Sciences) within the Poncelet
Laboratory [
http://

This study was partially supported by the French Space Agency (CNES).

Marked point processes are useful for image processing problems such as object extraction from remote sensing images. The advantage of these processes is that they allow the inclusion of
strong geometrical constraints on the objects to be detected. In our applications, a marked point process is defined by a density function with respect to the Poisson measure. Within the
framework of Gibbs point process, this density is expressed as a combination of several energy terms: firstly, a data energy term, which controls the localization of the objects with respect
to the data; secondly, prior information about the objects is given by internal energy terms corresponding to geometrical constraints on the objects. The weights associated with each internal
energy are the so-called `hyperparameters'. These `hyperparameters' used to be calibrated by hand. As their values depend on the kind of images studied, the calibration step is often long. In
order to develop fully unsupervised detection procedures, an estimation of these hyperparameters has to be performed. First of all, an estimation strategy is studied in the case of complete
data, for which the configuration,
*i.e.*the set of marked points corresponding to the objects, is known. In this case, several estimators such as maximum likelihood or pseudo-likelihood estimators may be derived.
However, since the normalizing constant of the process density is not tractable, maximum likelihood estimators cannot be derived directly. The likelihood is, rather, computed using an
importance sampling method. This implies the simulation of importance weights using MCMC methods. MCMCML estimators are then numerically obtained by maximizing the resulting estimated
likelihood. Another estimation method is based on the pseudo-likelihood, which is a combination of valid likelihoods associated with conditional events. Pseudo-likelihood estimators are then
obtained by maximizing the corresponding pseudo-likelihood. The interest of this last procedure is that one avoids the simulation step, since the normalizing constant does not appear in the
pseudo-likelihood. Of course, these estimators are sub-optimal, and their performance has to be compared to MCMCML estimators. Finally, in the more general framework of missing data,
*i.e.*when the configuration is unknown, there is in general no tractable closed form expression for the density associated with the observations. To tackle this problem, estimation
methods such as the Expectation-Maximization (EM) can be used. This work is in progress.

This study was partially supported by the COLORS Flamants project and was conducted in collaboration with Arnaud Béchet from La Tour du Valat [
http://

This work addresses the problem of detecting and counting breeding greater flamingos in aerial images of their colonies . We consider a stochastic approach based on object processes, also called marked point processes. Here, the objects represent flamingos, which are represented by ellipses. The density associated with the ellipse process is defined with respect to the Poisson measure. Thus, the problem is reduced to energy minimization, where the energy is composed of a regularizing term (prior density), which introduces some constraints on the objects and their interactions, and a data term, which links the objects to the features to be extracted in the image. Then, we sample the process to extract the configuration of objects minimizing the energy by a new and fast birth-and-death dynamics , leading to the total number of birds. This approach gives counts with good precision when compared to manual counts. Additionally, this approach does not need image pre-processing or supervision of the extraction, thus considerably reducing the overall processing time required to get the count. The algorithms were tested on images provided by the Tour du Valat. Two of them are presented in figure .

This Ph.D. was funded by an MESR grant and by INRIA/FSU Associated Team `SHAPES' [
http://

The goal of this research work is to develop a generic model for extracting geometric shapes from an image . The first applications we address concern tree crown and building extraction. The challenge is to find an appropriate shape representation and then a measure of shape similarity on the so-called shape space. This will allow us to define statistics on the shape space and then to sample shapes. The resulting statistical models will form part of the prior distribution of a probabilistic model for object extraction.

This year, we have studied a new representation, the so-called `q-representation'. This representation, in contrast to previous representations defined by the angle function, allows us to take into account not only shape bending but also shape compression and stretching in constructing geodesics between shapes.

Figures and show some examples in which one can see the difference between the geodesics for tree crown and building shapes constructed using the different approaches. Looking at the shapes along the geodesics, one notices that the branches of the tree crowns (Fig. ) and the angles of the buildings (Fig. ) are better preserved along the geodesic path using the new shape representation. This property is important for the calculation of the shape statistics, in particular the average shape, which will be more representative of the shape class.

This Ph.D. is co-supervised by Marc-Pierrot Deseilligny, chief scientist of the technical management of French National Geographic Institute (IGN). The data (satellite images of urban areas) were provided by French Space Agency (CNES).

Three dimensional models of urban areas are very useful for many kinds of applications, for example urban planning, radiowave reachability tests for wireless communications, and disaster recovery. However, 3D building reconstruction is a difficult problem, mainly due to the complexity of urban scenes.

We have developed a method for the 3D reconstruction of buildings from satellite images based on a stochastic approach . It consists in reconstructing buildings by assembling simple urban structures extracted from a library of 3D parametric models, rather like the toy Lego . Such a method is particularly well adapted to data of average quality such as satellite images. The approach is based on a density formulation defined within a Bayesian framework. The configuration that maximizes this density is found using an RJMCMC sampler, which solves the multiple parametric object recognition problem. Figure presents results on typical French town centres.

This Ph.D. is partially funded by a DGA/MRIS grant.

This work addresses the problem of target change detection in high resolution remotely sensed images . Its particularity is to take advantage of the data and differences available in image pairs of the same area taken at different times, rather than single images. In our previous work, we performed a first rough detection of changed areas based on an iterative Principal Component Analysis (PCA). This was followed by clustering based on an Entropy K-means algorithm. Both were based on radiometric data, although spatial dependence was introduced through a Markov random field regularization. This year, the research work has involved designing a polygonal approximation algorithm in order to extract geometrical data from the previously clustered zones. This algorithm is designed to minimize the Lebesgue measure of the symmetric difference between a contour and its Polygonal Approximation. As figure shows, this criterion is robust to outliers of the detected shape. Figure shows the convergence result on an aeroplane shape. Figure shows how building shadows (which qualify as changes) present particular directions that are eventually matched after algorithm convergence. Once obtained, the orientations of all the segments are classified in order to find a new criterion for discriminating changes (shadows and city street directions). Work in progress includes creating a model merging both geometric and radiometric data for the detected objects. Then a connectivity model defining relationships between the objects should improve the classification.

This Ph.D. was partly funded by SILOGIC and INRIA. We particularly thank Commandant Poppi (Fire brigade member and director of the cartography service, SDIS83 Draguignan) for interesting discussions.

Several studies have shown the effectiveness of using several coarse-resolution images to detect burnt areas. But these methods require at least two satellite images which are expensive. This work addresses the problem of burnt area mapping after forest fires from a single high-resolution post-fire image. It consists of delineating burnt areas from the radiometric information given by the different sensors of SPOT 5.

To discriminate burnt from unburnt areas, we use Support Vector Machines (SVM) , a supervised learning algorithm which provides high classification accuracy and good generalization capacity. Finally, we improve the classification by regularizing it with either a classical Markov model or mathematical morphology techniques.

SVM requires a set of observations (
*i.e.*a labelled training set) to predict the classification of unlabelled samples. To avoid the manual selection of this training set, we have proposed an automatic selection process
combining an entropy K-means algorithm and SVM
.

The classification methods (the simple SVM and the combination of K-means and SVM) are applied to several SPOT 5 images of Southern France (PACA and Corsica regions) containing various types of vegetation to test their efficiency , and are compared with classical algorithms such as K-means and K-Nearest neighbours . The extracted burnt areas are also compared to the corresponding ground truth provided by CNES, Infoterra-ESA, ONF-AM, and the SDIS3 and SDIS2B.

This work was supported by and performed as part of INRIA Associate Team `Shapes' in collaboration with Prof. Srivastava of Florida State University [
http://

The Fisher-Rao (FR) metric is the unique metric on spaces of probability measures to be invariant to push-forward by Markov mappings, and in particular, diffeomorphisms . Its expression in terms of half-densities is particularly simple: it is Euclidean, revealing that in the FR metric, spaces of probability measures are orthants of the unit sphere in spaces of measures. This enables the analytical computation of geodesics between probability measures, which previously required numerical computation.

The space of 1D diffeomorphisms of
[0, 1],
Diff ([0, 1]), is isomorphic to the space of probability densities on
[0, 1], and the FR metric can thus be used to measure distances between such diffeomorphisms. This enables the creation of `Gaussian' probability
distributions on
Diff ([0, 1]). Different samplings of curves can be described by the action of
Diff ([0, 1])on a fixed sampling (
*e.g.*uniform), and probability distributions on
Diff ([0, 1])can thus be pushed forward to the space of samplings.

Certain types of imaging modality or the outputs of certain types of image processing (
*e.g.*edge detection) lead to images that are essentially point sets, where some of the points correspond to the border of an object, while others are `noise'. Combining probability
distributions on shapes with probability distributions on samplings of curves, and an image formation model, enables the classification of objects from such images, as illustrated in
figure
, and described in
.

This work was supported by and performed as part of INRIA Associate Team `Shapes' in collaboration with Prof. Srivastava, Prof. E. Klassen, and S. Joshi of Florida State University [
http://

The Fisher-Rao metric is defined on spaces of densities, and hence pushes forward to
Diff ([0, 1]). These can be viewed as curves in one dimension. The question arises as to whether there is a natural generalization to spaces of
vector-valued densities, or equivalently to curves in higher dimensions: the derivative of a parameterized curve transforms as a vector-valued density under the action of
Diff ([0, 1]). There is in fact a one-parameter family of generalizations, and these metrics can be used to compute distances and geodesics between
parameterized curves, or projected to compute distances between `shapes',
*i.e.*parameterized curves modulo
Diff ([0, 1]), translation, rotations, scalings, etc. In 2D, all the members of the family are flat, and Euclidean coordinates can be found. For higher
dimensions, all members of the family are conformally flat, but only for one value of the parameter does the metric have zero curvature. Euclidean coordinates can be found, and these
facilitate the computation of geodesics between shapes in any number of dimensions. This also opens the way to computing distances and geodesics between `decorated' shapes,
*i.e.*with texture features attached to each point of the curve. The first row of figure
shows a geodesic between two 3D curves from two points of view, while the second row shows the results of clustering
American Sign Language shapes using geodesic distances computed from this metric. Further details are available in
,
.

This work was partially funded by CS-Toulouse.

The problem of reconstructing an image from a random set of irregular samples is a problem of great interest in various domain such as satellite imaging. In this work, we are interested in
solving a problem of image restoration with different aspects: reconstruction from irregular samples, deconvolution, and denoising. The context is a satellite stereoscopic acquisition of a
scene. Thus we have two regular acquisitions of the same scene. By applying the disparities between the two images to the reference image, we get an irregularly sampled new image which should
be identical to the second image of the stereopsis pair (apart from some details due to moving objects during the time between the acquisitions of the stereoscopic pair). As a matter of fact,
the second image can be considered as an irregularly sampled acquisition (in comparison to the reference image) and the problem of reconstructing the reference image from the second image
knowing the disparities between the two images can be considered as an irregular sampling problem. We also consider at the same time the deconvolution problem, by considering the point spread
funstion (PSF) of the acquisition system. The noise considered here has two different parts: the first is due to the acquisition system, and can be considered to be white Gaussian noise,
while the second is due to errors in the computation of the disparities between the two images of the stereoscopic pair. Such an error may have disastrous effects in urban images: poor
estimation of the position of a sample located on the top of a building may place this sample in a place where there should be some shadow. As a matter of fact, some errors in the estimation
of irregular samples may completely change the value of a pixel, which can be seen as impulsive noise. We reconstruct and restore the image by minimizing a functional with an
l^{1}-norm on the data term in order to take into account the impulsive noise and a standard TV regularizing term. We use a smoothed approximation of the
l^{1}-norm around zero, and we minimize the functional using a gradient descent algorithm. We have shown that using an
l^{1}-norm on the data term is more robust to impulsive noise (see figure
).

This work was partially funded by Astrium/EADS Toulouse.

The problem addressed in this work is the reconstruction of a high resolution image from several low resolution images. In the context of remotely sensed images, this can be applied to
geostationary satellites, for example, whose resolution is low due to the high altitude of the satellite. We assume that we know the shift between the low resolution observed images. At the
same time, we address the problems of noise removal and deconvolution in order to remove the effect of the point spread function (PSF) of the acquisition system. The problem is formulated as
an inverse problem, and we regularize it by minimizing a functional composed of the data term and a TV regularization term. In order to be robust to small errors in the shift estimation,
which can induce high intensity errors at borders of objects, and which can be viewed as impulsive noise, we propose to minimize a functional with an
l^{1}-norm in the data term to be increase robustness. As the
l^{1}-norm is not differentiable at zero, we use a smooth approximation of it for both terms, and minimize the functional by using a gradient descent procedure. Tests have been made on
images provided by Astrium/EADS, comparing the results obtained with the
l^{2}-norm and the
l^{1}-norm in the data term, as well as comparing to other recent methods in the literature (see figure
). When there are no errors in the shift, we can use the functional with the
l^{2}-norm. In this case, very fast algorithms can be derived. In the other case, faster algorithms are still to be developed.

This Ph.D. is co-supervised by Gilles Aubert, professor of the J.-A. Dieudonné Mathematics Laboratory of the University of Nice Sophia Antipolis [
http://

We have developed new algorithms
to minimize total variation and more generally
l^{1}-norms under a general convex constraint. The algorithms
are based on a recent advance in convex optimization proposed by Yurii Nesterov. Depending on the regularity of
the data fidelity term, we solve either a primal problem or a dual problem.

First, we show that standard first-order schemes allow solutions of precision in iterations at worst. For a general convex constraint, we propose a scheme that allow solutions of precision in iterations. For a strongly convex constraint, we solve a dual problem with a scheme that requires iterations to obtain a solution of precision .

Thus, depending on the regularity of the data term, we gain from one to two orders of magnitude in the theoretical convergence rates with respect to standard schemes. We perform some numerical experiments which confirm the theoretical results for various problems. Compared to standard optimization procedures, computing times are reduced by a factor ranging from 4 to 20.

This Ph.D. is co-supervised by Baogang Hu, from LIAMA/CASA, Chinese Academy of Sciences [
http://

The problem addressed in this work is the extraction of the region containing the road network from very high resolution ( m) satellite images. In our initial work , a phase field higher-order active contour was used to model the network region and its relation to the image, and thus to extract the main road network in dense urban areas from a single QuickBird panchromatic image (see figure ). At 1/8resolution, the complete main road network is successfully retrieved (see figure ). However, at the original resolution, the detail in the image results in errors along the boundary of the roads and in the detection of spurious regions in the background (see figure ).

The prior energy used was
*generic*: it incorporates constraints on the form of the road network region that are true of any road network. To improve the results at the original resolution, we developed a
*specific*prior energy, linking the extracted road network region to a GIS map of the road network
. The GIS map was obtained a few years earlier than the satellite images, and thus represents a slightly
different road network. To test the method, we introduced further artificial errors in order to increase the difficulty of the problem (see figure
). The additional prior energy measures the difference between the network region in the GIS map and the extracted network.
Its main effect is to reduce false detections in the background. Experiments show that at the original resolution, our model is able to keep the unchanged roads, to correct the mistakes, and
to extract new roads (see figure
). In order to free the method from the need for a GIS map, we replaced the GIS map by the result obtained with our previous
method at a lower resolution. The results show little change, showing that a GIS map, although useful, is not necessary.

We then turned to the extraction of secondary roads. These are far harder to deal with because their radiometric properties are very similar to those of the background, and because small roads are often occluded by shadows and trees. To deal with these difficulties, we developed a new nonlinear nonlocal prior energy term. This is a new type of higher-order active contour/phase field term. It allows the interaction between points on the same side of a road to be stronger and of longer range than the interaction between points on opposite sides of a road. The preliminary results at 1/4and 1/2resolution are extremely promising, clearly outperforming the old model (see figure ).

This Ph.D. is co-supervised by Zoltan Kato, Assistant Professor at the University of Szeged, Hungary. The data (aerial images of French forests) were provided by the French National Forest Inventory (IFN).

In this work, we propose a phase field version , of our earlier higher-order active contour (HOAC) `gas of circles' model , , and develop a multispectral data term for colour infrared aerial images .

The HOAC `gas of circles' model is an effective tool for modelling circular shapes. Nevertheless there are some difficulties. It is complicated to express the space of regions in the contour representation, and consequently difficult to work with a probabilistic formulation. In addition, from the algorithmic point of view, the current model does not allow enough topological freedom, and the implementation of the HOAC model is difficult and computationally expensive. However, it is possible to create an alternative formulation of HOAC models, based on the `phase field' framework much used in physics to model regions and interfaces . In order to make use of the stability analyses developed for the HOAC models, we computed, as a function of the HOAC energy parameters, the phase field energy parameters that produce an equivalent model. This means that we can adjust the phase field parameters to ensure stable circles of a given radius . We extended the phase field `gas of circles' model to the case of an inflection point rather than a minimum in the circle energy at the desired radius . The use of the phase field framework cuts execution times by one or two orders of magnitude.

We have also introduced a data model using all three bands of the colour infrared (CIR) images . We studied the quality of the extraction results produced by modelling the three bands as independent or as correlated. We have shown that, even at the level of maximum likelihood, the inclusion of `colour' information, and in particular, interband correlations, can improve the results, and in conjunction with the region prior, the full model is considerably better than that based on one band alone. We use the models to extract tree crowns from aerial images, but the models are not restricted to forest management: they can be applied to the detection of any circular objects.

Figure (a) shows a CIR image of a plantation, with a field in its upper part, while figure (b) shows the segmentation result obtained using the multispectral `gas of circles' model. Figure (c) shows a panchromatic aerial image with sparsely planted poplars, while figure (d) shows the result obtained using the phase field `gas of circles' model.

This work was partially funded by EU Network of Excellence MUSCLE.

Higher-order active contours (HOACs) are models of regions containing sophisticated prior geometric knowledge about region shape introduced via long-range interactions between region boundary points , . They have been used to extract road networks , and tree crowns , from remote sensing images using models of network shapes and a `gas of circles' respectively.

The shapes modelled by a given energy can vary considerably with the model parameters. In particular, the same energy is used for the two applications mentioned above, only with different
parameter values. In order to set the parameters of the model for a given application, we therefore need to know which shapes are modelled in which parameter ranges. The aim of this work is
thus to determine the `phase diagram' of the basic HOAC model,
*i.e.*to determine the parameter ranges leading to stable circles and/or stable bars (a bar is a simplified network).

The stability analysis for both configurations (bar and circle) is based on a functional Taylor series expansion of the energy up to second order around the shape whose stability we wish to analyse. The first functional derivative must be zero and the second functional derivative non-negative definite in order for the shape to be stable, which places strong constraints on the model parameters, allowing us to reduce the effective dimension of the parameter space, in some cases to one dimension. Figure shows two pieces of the phase diagram resulting from these analyses, corresponding to circles (left) and bars (right). The coloured zones of the diagrams represent the parameter ranges leading to stable circles or stable bars respectively. The upper part of each zone corresponds to circles with negative energy per circle, or negative energy per unit length of bar, while the lower parts correspond to positive energies. This division is important as it has a dramatic influence on the behaviour of the model: if the energy of a circle is negative then as many circles as possible will be created, while if the energy per unit length of a bar is negative then the bar will lengthen without limit. The corresponding Gibbs distributions are then not normalizable. This work will be published in RFIA 2008.

This study is part of a Ph.D. funded by the University `Federico II' of Naples, Italy, and was partially supported by EU Network of Excellence MUSCLE, and conducted in collaboration with Prof. G. Scarpa (University of Naples).

This work addresses the problem of land classification via pixel-based image segmentation of low- and mid-resolution multispectral satellite images. Unsupervised segmentation is here pursued using a particular class of Markov random fields, namely tree structured MRFs (TS-MRFs). These models allow the hierarchical representation of a 2D field by the use of a sequence of nested MRFs, each corresponding to an internal node of the tree. Consequently, the segmentation algorithm can perform optimization of such models by recursively working on a single internal node at a time, from the root to the leaves, with a significant reduction in complexity.

In our former work on TS-MRFs, only binary tree structures were used, but such a constraint can be too tight when attempting to fit arbitrarily structured data within a hierarchical tree representation. In this work, we propose a variation to our unsupervised segmentation algorithm that allows the use of generic tree structures, providing dynamic dimensionality selection for MRFs located at each node of the tree, and therefore a new structure evolution strategy, based on Mean-Shift analysis applied to the joint spatial/spectral feature space. Mean-shift is also used to provide initial conditions for the optimization of each MRF.

This work is being done as part of the ACI MULTIM [
http://

An important evolution in image processing is the passage from scalar data to multi-channel data, with the development of multi- and hyper-spectral cameras, particularly for astronomical
and satellite imaging. A crucial problem in this context is the restoration of these data,
*i.e.*the removal of degradations introduced during the acquisition procedure. An important aspect is to take into account the dependency of the channels in the restoration procedure.
The aim of this ACI is to develop mathematical models for multi-spectral image restoration, mainly using variational and PDE multidimensional methods.

This Ph.D. is co-supervised by Henri Maître, deputy director of École Nationale Supérieure des Télécommunications (ENST), Paris [
http://

Remote sensing image databases have grown enormously with recent advances in image acquisition technologies. These images offer a huge amount of potentially useful information, but retrieving this information from such large volumes of data using indices of relevance to applications is an extremely difficult problem. In this work, we take a novel approach to this problem. Instead of using correlations between low-level image features (representing short-range pixel dependencies) and semantic content to retrieve images, we use correlations between one type of semantic content (representing long-range pixel dependencies) and another. Specifically, we use the fact that the properties of road networks vary from one type of geographical environment to another to identify and retrieve these environments. This presupposes that we can find automatically the road network in an image; we test various methods for this task in order to evaluate the robustness of the method to errors in the network. To describe the properties of road networks, we use various statistics computed from a graphical representation of the extracted network to quantify properties such as density, curvature, spatial variation, and so on . In order to deal with the failure of the extraction methods in dense urban areas, a second set of features computed from a segmentation of these areas is used . We use the extracted features to classify large satellite images at a coarse scale using a one-vs-rest probabilistic Gaussian kernel support vector machine. An sample result is shown in figure .

This research was partly supported by the P2R Franco-Israeli Collaborative Program [
http://

In this work, we propose an iterative algorithm for 3D confocal microcopy image restoration. The image quality is limited by the diffraction-limited nature of the optical system, which causes blur, and the reduced amount of light detected by the photomultiplier leading to noise having Poisson statistics. Wavelets have proved to be very effective in image processing and have gained much popularity. Indeed, they allow the efficient denoising of images by applying a thresholding to the coefficients. Moreover, they are used in algorithms as a regularizing term and seem to preserve textures and small objects.

We propose here a 3D iterative wavelet-based algorithm and make some comparisons with state-of-the-art methods for restoration. More precisely, we compare three methods:

Method 1: denoising (wavelet thresholding) then deconvolution (Richardson-Lucy);

Method 2: deconvolution (PSF inversion) then denoising (estimation of the noise and then wavelet thresholding);

Method 3: combine denoising and deconvolution.

We plan to use pre-processing similar to the Anscombe transform or the Fisz transform in order to stabilize the noise variance and consequently apply more conventional methods developed for Gaussian noise.

Different tests were made on synthetic images, real images, and phantom images (
*i.e.*the object put under the microscope is known). The latter class of images allows the numerical and visual evaluation of the methods. Figure
shows a result on a phantom image of a shell. The objective here is to recover the initial shell dimension. We know that the
thickness lies between
0.5and
0.7
m.

Looking at figure , method 3 seems to lead to the best results. If we look at the intensity profiles of the images (see figure ), we can see that method 3 allows a more accurate recovery of the real thickness of the shell.

This Ph.D. was funded by an INRIA CORDI Fellowship. It forms part of the P2R Franco-Israeli project [
http://

This work addresses the problem of blind deconvolution for fluorescence microscopy. An iterative method can restore fluorescence Confocal Laser Scanning Microscopy (CLSM) images and parametrically estimate the acquisition system point spread function (PSF). The CLSM is an optical fluorescence microscope that scans a specimen in 3D and uses a pinhole to reject most of the out-of-focus light. However, the quality of the images suffers from two basic physical limitations. The diffraction-limited nature of the optical system and the reduced amount of light detected by the photomultiplier cause blur and photon counting noise respectively. These images can hence benefit from post-processing restoration methods based on deconvolution. An efficient method for parametric blind image deconvolution involves the simultaneous estimation of the specimen 3D distribution of fluorescent sources and the microscope PSF . By using a model for the microscope image acquisition physical process, we reduce the number of free parameters describing the PSF and introduce constraints. The parameters of the PSF may vary during the course of experimentation, and so they have to be estimated directly from the observed data. A prior model of the specimen is further applied to stabilize an alternate minimization algorithm and to converge to the solution.

The model was previously tested on 2D images and the restoration process was very effective. However, extending it to 3D images poses a new challenge because the sampling is different for the radial and the axial plane. Another important obstacle is that the amount of spread in the intensity is higher in the Z direction than in the XY plane. Hence direct extension from 2D to 3D case is not obvious. Additional changes to the parametric estimation algorithm ensured a more robust estimation of the parameters of the PSF without imposing any constraints on the estimation step size. The results obtained on some 3D synthetic images are as shown in figure .

This Ph.D. is co-supervised by Gilles Aubert, professor of the J.-A. Dieudonné Mathematics Laboratory of the University of Nice Sophia Antipolis [
http://

We propose a new model to detect locally thin filaments in 2D or 3D confocal microscopy images
,
. This model uses an approximation of the PSF by a Gaussian function
and take into consideration the effect of the discretization due to the CCD sensors and additive Gaussian noise
b.

First, we show that the observed image of a linear filament passing through a point
xwith an intensity
Aand a direction
vat pixel
y_{i}, without additive Gaussian noise
b, is given by:

We look for pairs
(
F,
b)where
Fis a filament with parameters
(
x,
V
v,
A)compatible with the model
above, and b is a Gaussian with a minimum
L^{2}norm. We use a gradient descent method in order to obtain these pairs.

The examples in figure show the results of the proposed method on a synthetic 2D image.

Extraction and characterization of line networks in satellite images for retrieval from image databases. Contract #293

Parameter estimation for 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.

Satellite image reconstruction from irregular sampling. Contract #2104.

Super-resolution for satellite imaging. Contract #2402.

Evaluation of the damage after a forest fire from high resolution satellite images. Contract #1156.

Road network updating in dense urban areas from very high resolution optical images. Contract #1675.

Semi-automatic methods for forestry cartography using aerial and high resolution satellite images. Contract #1467.

Target detection through texture perturbation analysis. Grant from the DGA and CNRS.

Tree detection and counting in orchards. Contract #2576.

Urban area extraction from SPOT5 images. Contract #2713.

Detection of flamingos in aerial images. In collaboration with the Tour du Valat (Sansouire Foundation) in Arles (Arnaud Béchet) (
http://

In collaboration with the J.-A. Dieudonné Laboratory of CNRS/UNSA (G. Aubert, L. Almeida), the Pasteur Institute in Paris (J-C Olivo-Marin, C. Zimmer), SAGEM DS at Argenteuil (D. Duclos, Y. Le Guilloux), and ENS Cachan (J.F. Aujol).

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

In collaboration with Paris 6 (P. Combettes (PI)), Paris 5 (L. Moisan), Ecole Polytechnique (A. Chambolle), J.-A. Dieudonné Laboratory of CNRS/UNSA (G. Aubert), University of Paris
Est-Marne La Vallée (JC. Pesquet), Observatoire Midi-Pyrénées (S. Roques). Website:
http://

In collaboration with the Signal and Image Processing laboratory of ENST (M. Campedel, Y. Kyrgyzov, B. Luo, H. Maître, M. Roux (PI)), INRIA project-team IMEDIA (O. Besbes, S. Boughorbel,
N. Boujemaa, M. Crucianu, M. Ferecatu, V. Gouet) and URISA of Sup'Com Tunis (Z. Belhadj, A. Ben Azza, R. Tebourbi). Website:
http://

The Ariana project-team is a participant in the European Union Sixth Framework Network of Excellence MUSCLE (Multimedia Understanding through Semantics, Computation and Learning), contract
FP6-507752, in collaboration with 41 other participants around Europe, including four other INRIA project-teams. Web site:
http://

In collaboration with the Pasteur Institut (Bo Zhang, Jean-Christophe Olivo-Marin [PI]), the Weizmann Institute (Zvi Kam) and Technion (Arie Feuer [PI]). Website:
http://

In collaboration with the Vision Group of Florida State University (A. Srivastava (PI), V. Patrangenaru, S. H. Joshi, and W. Liu ). Website:
http://

In collaboration with LIAMA (V. Prinet) and the VISTA/IRISA project-team (J.F. Yao, P. Bouthemy).

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

Members of the Ariana project-team participated in the Fête de la Science at EAI CERAM in October.

The Ariana project-team organized numerous seminars in image processing during 2007. Eighteen researchers were invited from the following countries: Belgium, Canada, China, the Czech Republic, France, Hungary, Ireland, Israel, the Netherlands, Spain, Switzerland, the UK, and the USA. 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 (École Polytechnique, ENS Cachan, ENPC, Sup'Aéro) and from John Hopkins University, USA; helped students of the Classes Préparatoires with TIPE in France; and gave information on remote sensing image processing to high school students in Mauritius.

The work of the Ariana project-team was reported in Interstices (special topic on tree detection in aerial images), in Code Source for the 40
^{th}anniversary of INRIA, and in LISA in May (special issue dedicated to the environment).

The Ariana project-team presented demos to visitors from the Association Aéronautique et Aérospatiale Francaise (AAAF) in March.

All the members of the Ariana project-team participated in a `brain-storming' session in Sospel in June.

Avik Bhattacharya gave a cross-seminar (Ariana/Maestro) in May at INRIA Sophia Antipolis. He presented a paper at the Fifth International Workshop on Content-Based Multimedia Indexing (CBMI 2007), Bordeaux. He presented a poster at the 1 Symposium of the CNES/DLR/ENST Competence Centre on Information Extraction and Image Understanding for Earth Observation, Paris, in June, and an oral presentation at the DLR centre in Oberpfaffenhofen in March. He gave a talk at the ACI Masses de Données QuerySat meeting at ENST, Paris.

Alexis Baudour presented a poster at SSVM, Ischia, Italy in May. He gave a talk at SMAI, Praz-sur-Arly, France, in June. He presented a poster at GRETSI, Troyes, France, in September.

Caroline Chaux presented a paper at the SPIE Wavelet XII conference in August in San Diego, California, USA. She presented a seminar at CEA Saclay in September.

Stig Descamps gave a seminar for the GdR ISIS meeting in July at Télécom Paris, France.

Aymen El Ghoul gave a seminar at LIAMA in July in Beijing, China. He presented his Masters project in June at the Ecole Supérieure des Télécommunications de Tunis, Tunisia.

Alexandre Fournier and Olivier Zammit attended the workshop `Utilisation des drones pour la prévention et la lutte contre les feux de forêt' organized by the `Pôle de Compétitivité Pegase', and the `Pôle Risques', in May in Gardanne, France.

Alexandre Fournier presented a paper at the TAIMA conference in Hammamet, Tunisia, and gave a seminar at Sup'Com in Tunis, Tunisia, both in May. He gave a seminar at the CNES-DLR workshop in March in Oberpfaffenhofen, Germany. He also presented the work of the Ariana project-team in Spanish to an official delegation from Valencia, Spain.

Peter Horvath gave a cross-seminar (Ariana/Odyssée) in May at INRIA Sophia Antipolis. He presented his work at ACIVS in August in Delft, the Netherlands, and at EUSIPCO in September in Poznan, Poland. He gave a talk at University of Szeged in Hungary in September and at ETH Zurich in November.

Maria Kulikova presented a poster at the `Shape Day' workshop in the Department of Statistics of Florida State University, Tallahassee, Florida, USA in April. She participated at The International Summer School on Pattern Recognition in July, where she also presented a poster. She presented a paper at EUSIPCO, Poznan, Poland in September.

Florent Lafarge gave talks at the “Atelier PNTS: la très haute résolution spatiale en télédétection” in September in Nantes, France, and at the EURANDOM workshop on image analysis and inverse problems in Eindhoven, The Netherlands. He presented papers at ICIP in San Antonio, USA, at EUSIPCO in Poznan, Poland, and at PIA in Munchen, Germany, all in September.

Praveen Pankajakshan presented a paper at the International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) in August in Lyon, France. He presented his work at the first P2R Franco-Israeli joint meeting in February, at the Pasteur Institute, Paris, France. He presented his work at the second P2R Franco-Israeli joint meeting in October, at Sophia Antipolis and gave a cross-seminar at ED STIC in November.

Ting Peng participated in the Summer School on Statistical Learning, Microsoft Research Asia, Beijing, China, in July. She participated in EMMCVPR in EZhou, Hubei, China, in August. She presented a paper at BMVC, Warwick, UK, in September. She participated in the THRS TU Workshop of PNTS, Nantes, France, in September.

Pierre Weiss, presented a poster at CODE in April in Paris, France. He gave a talk at SMAI, Praz-sur-Arly, France, in June. He gave a talk at GRETSI, Troyes, France, in September.

Olivier Zammit made a presentation at the INTECH seminar in June in Sophia Antipolis, France. He presented a paper at IGARSS in July in Barcelona, Spain. He presented another paper at the GRETSI Colloquium in September in Troyes, France. He attended the `Observation et télédétection pour une meilleure gestion des risques' conference organized by POPSUD in March in Aix en Provence, France, and the Drones conference organized by the `Pôle de Compétitivité Pegase' and the `Pôle Risques' in Valabre in May.

Laure Blanc-Féraud gave an invited talk at the Workshop `An interdisciplinary approach to Textures and Natural Images Processing' in January in Paris, France. She gave a seminar at the Pierre et Marie Curie University in April in Paris, France. She gave an invited tutorial talk at the ORASIS conference in June in Obernai, France. She attended the conference SSVM in May in Italy and GRETSI, Troyes, France, in September. She participated in the P2R meetings at the Pasteur Institute in Paris in February and at Sophia Antipolis in October. She organized and attended ANR meetings, one in Paris in March and one in Nice in September.

Xavier Descombes gave an invited talk at the Workshop `An interdisciplinary approach to Textures and Natural Images Processing' in January in Paris, France. He visited the Department of Statistics of Florida State University for a week in April, and gave an invited talk as part of the `Shape Day' workshop organized at FSU. He gave a seminar at IITP in Moscow in November. He presented a paper at VISAPP in Barcelona in March. He took part in two ANR `Micro-Réseaux' meetings. He gave a seminar organized by the Geometrica project-team at INRIA Sophia Antipolis. He took part in two meetings with CNES where he gave talks.

Ian Jermyn gave an invited talk at CNES in Paris as part of the ORFEO Accompaniment Programme day in January. In March, he took part in a meeting with CNES, at which he gave a talk, and also gave a seminar to the INRIA APICS project-team. He visited the Department of Statistics of Florida State University for a week in April, and gave an invited talk as part of the `Shape Day' workshop organized at FSU. Also in April, he gave a paper at URBAN in Paris. In May, he presented his work to a representative of the European Commission Joint Research Centre at INRIA Sophia Antipolis. In June, he attended a meeting of the CNES/DLR/ENST Centre of Competence in Paris. In July, he attended a meeting of ACI QuerySat in Paris, and gave a talk; he attended the Journée Pôle IMA-SIG-SYS and gave a talk; and he presented his work to the company CS. In August, he gave a paper at CAIP in Vienna. In September, he took part in a meeting with CNES in Toulouse, and gave a talk. In November, he visited the University of Szeged, Hungary for a week; he attended the COST GTRI Equipe Associées evaluation meeting in Paris; and he took part in a plenary meeting of the MUSCLE Network of Excellence, at which he gave a talk. In December, he took part in a meeting with the management of Total.

Josiane Zerubia made presentations of the work of the Ariana project-team in January to Astrium/EADS and Infoterra France in Toulouse, and to the Director of the French Forest Inventory (IFN) in Sophia Antipolis. In May, she presented Ariana's work to a representative of the European Commission Joint Research Centre in Sophia Antipolis and to the CEO of Noveltis in Toulouse. Also in May, she helped evaluate CNES as a member of the CERT, attended the GMES day organized by CNES in Toulouse, and presented Ariana's work to the Director of ONERA in Salon de Provence in relation with the `Pôle de Compétitivité Pegase'. In June, she gave a seminar on risk management as part of INTECH day, and organized the Ariana Brain Storming meeting, which took place over three days in Sospel. In September, she presented Ariana's work at CNES in Toulouse, and visited Sanofi Aventis and presented a paper at EUSIPCO, Poznan, Poland. In October, she organized a one-day international workshop on confocal and wide-field microscopy at the Hotel Mercure in Sophia Antipolis, as well as the P2R Franco-Israeli meeting on confocal and wide-field microscopy in Sophia Antipolis. In November, she presented Ariana's work at Silogic in Toulouse and to the Rotary Club of Antibes Juan les Pins, visited Sanofi Aventis for a second time, and attended the MUSCLE EU Network of Excellence plenary meeting in Paris. In December, she organized the visit to the Ariana project-team of a Hungarian delegation from the University of Szeged and SZTAKI, and a delegation from Total management.

C. Chaux was a reviewer for IEEE TCAS, IEEE TSP, and IEEE TIP.

P. Pankajakshan was a reviewer for the SPIE journal on Optical Engineering.

P. Weiss was a reviewer for SITIS'07.

Laure Blanc-Féraud is a regular reviewer for the journals IEEE TIP and IJCV, and for the conferences Eusipco'07, EMBS'07, GRETSI'07. She is reviewer for the ANR (programme blanc and jeunes chercheurs), she was a referee for the Scientific Commission of the EPFL (Swiss Federal Institute of Technology, Lausanne) for awarding Doctoral Fellowships in Imaging Technology for Biology and Medicine. She is a member of the jury for the Ph.D. award of Club EEA, Signal and Image section. She is a member of the INRIA Committee in charge of assigning Ph.D. and postdoctoral grants (CORDI programme). She was a reviewer for one Ph.D. thesis and one HdR, and attended one Ph.D. defence.

Xavier Descombes is a regular reviewer for the journals IEEE TIP, IEEE Medical Imaging, and IEEE TPAMI, and for IJCV. He was a reviewer for Gretsi'07, RFIA'07, ICCV'07 and CVPR'07. He is reviewer for the ANR (programme blanc and jeunes chercheurs). He was a reviewer for five Ph.D. theses and was on one Ph.D. committee. He is also an expert for the DDRT program on `Jeunes Entreprises Innovantes'.

Ian Jermyn was a regular reviewer for the IEEE TSP, IEEE TIP, and IEEE TPAMI, for the ISPRS Journal of Photogrammetry and Remote Sensing, and for the Journal of Mathematical Imaging and Vision. He was a reviewer for ECCOMAS'07, EUSIPCO'07, and ICIP'07. He was on two Ph.D. committees.

Josiane Zerubia was a reviewer for four Ph.D. theses and one HdR, and a member of three Ph.D. committee. She was a reviewer for IJCV, and SFPT (Revue Française de Photogrammétrie et de Télédétection). She was a reviewer or a program committee member for ICASSP'07, ISBI'07, ICIP'07, SPIE-ISPRS'07 (`Image and Signal Processing for Remote Sensing'), ICCV'07, MICCAI'07, SISP'07, GRETSI'07, and TAIMA'07.

Laure Blanc-Féraud was vice-director of the I3S Laboratory of CNRS and the University of Nice Sophia Antipolis until September. She is vice-director of the Gdr ISIS in charge of the imaging theme. She is a member of the GRETSI association and of its desk. She is a member of the scientific committee of the École Doctorale STIC of UNSA. She is a member of the organizing committee of the summer school Peyresq organized every year in July. She has organized four days of training on image processing for Sagem DS in Argenteuil in October. She is part of the ANR Scientific Committee for STIC (`Sciences et Technologies de l'Information et de la Communication').

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 also a member of the expert committee `Multi' of CNRS. He is computer systems coordinator for the Ariana project. He was coordinator of the collaboration COLORS 'Flamants' between Ariana and the Sansouire Foundation (`Station Biologique de la Tour Valat'). He is the Ariana coordinator for the ACI `QuerySat'.

Ian Jermyn is a member of the Comité de Suivi Doctoral at INRIA Sophia Antipolis, and, since October, of the COST Groupe de Travail Relations Internationales of INRIA. He is co-computer systems coordinator for the Ariana project. He is the coordinator of the Ariana project-team's efforts within the EU Network of Excellence MUSCLE. He was a member of the program committees for EUSIPCO'07 and ECCOMAS'07.

Josiane Zerubia is an IEEE Fellow. She is a member of the Biological Image and Signal Processing Technical Committee of the IEEE Signal Processing Society. She is an
Associate Editor of the collection `Foundation and Trends in Signal Processing' (
http://

Alexis Baudour was lab instructor for `Mathematics for digital images' (64h), at the IUT of Nice Sophia Antipolis.

Alexandre Fournier was an advisor for Image Processing projects (21h) at Poly'Tech Nice-Sophia Antipolis.

Maria Kulikova was teaching assistant for the course `Games and Strategies' (64h) at Poly'Tech Nice-Sophia Antipolis (UNSA).

Florent Lafarge was course instructor for `Image processing' (8h) at ENSG (Marne-La Vallée).

Pierre Weiss was lab instructor for `Digital signal processing' (32h), and for `Mathematics' (40h) at Poly'Tech Nice-Sophia Antipolis (UNSA).

Laure Blanc-Féraud taught at Sagem DS (6 hours) and at Poly'Tech Nice-Sophia Antipolis (6 hours).

Xavier Descombes taught at Sup'Aero (20h), at Poly'Tech Nice-Sophia Antipolis (UNSA) (9h) and at SAGEM DS (6h).

Ian Jermyn taught `Image analysis' (6h) at Poly'Tech Nice-Sophia Antipolis (UNSA), and `Filtering and segmentation of space imagery' (2.5h) at Sup'Aéro.

Josiane Zerubia taught the module `MRF models in image processing' for the Masters 2 course IGMMV at the University of Nice-Sophia Antipolis (3h). She was director of the course `Filtering and segmentation of space imagery' at Sup'Aéro (40h, of which 20h taught), where she also taught as part of the course `Variational methods for image processing' (2.5h). She also taught 3 hours at SAGEM DS.

Alexis Baudour: `Segmentation and deconvolution of 3D images', University of Nice-Sophia Antipolis. Defence expected in 2008.

Aymen El Ghoul: `Champs de Phase pour l'extraction de réseaux à partir d'images de télédétection', University of Nice-Sophia Antipolis. Defence expected in 2010.

Alexandre Fournier: 'Détection de cibles par une analyse des perturbations de la texture', École Nationale Supérieure de l'Aéronautique et de l'Éspace, Toulouse. Defence expected in 2008.

Maria S. Kulikova: `Reconnaissance de formes pour l'analyse de scènes', University of Nice-Sophia Antipolis. Defence expected in 2009.

Praveen Pankajakshan: `Blind biological image deconvolution', University of Nice-Sophia Antipolis. Defense expected in 2009.

Ting Peng: `Variational models for road network updating in dense urban areas from very high resolution optical images', CASIA, Chinese Academy of Sciences, Beijing, China and University of Nice-Sophia Antipolis. Defence expected in 2008.

Pierre Weiss: `Multispectral image processing with PDEs', University of Nice-Sophia Antipolis. Defence expected in 2008.

Olivier Zammit: `Forest fire damage assessment from satellite images', University of Nice-Sophia Antipolis. Defence expected in 2008.

Avik Bhattacharya: `Indexing of satellite images with structural information', École Nationale Supérieure des Télécommunications, Paris. Defended on December 14.

Peter Horvath: `The “gas of circles” model and its application to tree crown extraction', University of Szeged, Hungary, and University of Nice-Sophia Antipolis. Defended on December 3.

Florent Lafarge: 'Modèles stochastiques pour la reconstruction d'environnements urbains', École Nationale Supérieure des Mines, Paris. Defended on October 2.