Ariana/Maestro : Stochastic and graph methods for networking and image processing

Format des présentations : 30mn + questions

Avik Bhattacharya, projet Ariana :
"Image Retrieval from Road Network using a Graph Representation"
Résumé : Automatic interpretation of remote sensing (RS) images and the growing interest for query from large remote sensing image archives rely on the ability and robustness of information extraction from the observed data. Road network can be used to characterize and retrieve images from remote sensing database. Generally speaking, it can be used to locate geographical position of the region considered or to identify the image captor or to retrieve images with geometrical and topological properties. At the coarsest level, one can identify two major components of retrieval problem : representation and matching. At the onset all image information mining and retrieval paradigm require an appropriate extraction method to abridge the object from the image considered and it requires one common representation space. We represent the topology of roads as graphs, where special nodes represent extremities and junction and arcs (collection of nodes) represents roads.

Guillaume Perrin, projet Ariana :
"Stochastic Geometry for Forest Resource Assessment"
Résumé : High resolution aerial and satellite images of forests have a key role to play in natural resource management. As they enable forestry managers to study forests at the scale of trees, it is now possible to get a more accurate evaluation of the resources. Automatic algorithms are needed in that prospect to assist human operators in the exploitation of these data. In this talk, we present a stochastic geometry approach to extract 2D and 3D parameters of the trees, by modelling the stands as some realizations of a marked point process of ellipses or ellipsoids, whose points are the locations of the trees and marks their geometric features. As a result we obtain the number of stems, their position, and their size. This approach yields an energy minimization problem, where the energy embeds a regularization term (prior density), which introduces some interactions between the objects, and a data term, which links the objects to the features to be extracted, in 2D and 3D. Results are shown on Colour Infrared aerial images provided by the French National Forest Inventory (IFN).

Nicolas Bonneau, projet Maestro
"Random Matrices for Wireless Communications"
Résumé : Random matrix theory deals with the asymptotic behaviour of the eigenvalues (or other characteristics) of large random matrices. This field has been extensively studied since the fifties, but it is only recently that large system analysis based on random matrix theory was applied to wireless communication systems. In particular, it can be used to gain insight on CDMA (Code Division Multiple Access) cellular systems with large spreading factors and large number of users. CDMA is a popular multi-access technique enabling high rate wireless communications. With the help of random matrix theory, explicit formulas for performance metrics, depending only on a few meaningful parameters, can be derived in many cases of interest. Even if the results are obtained in the asymptotic regime, they give very accurate predictions of the system's behavior in the finite size case, as shown by simulations.

Natalia Osipova, projet Maestro
"Monte Carlo Methods for PageRank computation : When one iteration is sufficient"
Résumé : PageRank is one of the principle criteria according to which Google ranks Web pages. PageRank can be interpreted as a frequency of visiting a Web page by a random surfer and thus it reflects the popularity of a Web page. Google computes the PageRank using the power iteration method which requires about one week of intensive computations. In the present work we propose and analyze Monte Carlo type methods for the PageRank computation. There are several advantages of the probabilistic Monte Carlo methods over the deterministic power iteration method: Monte Carlo methods provide good estimation of the PageRank for relatively important pages already after one iteration; Monte Carlo methods have natural parallel implementation; and finally, Monte Carlo methods allow to perform continuous update of the PageRank as the structure of the Web changes.