Detecting skin lentigo in reflectance confocal microscopy images is an important and challenging problem. This imaging modality has not yet been widely investigated for this problem and there are a few automatic processing techniques. They are mostly based on machine learning approaches and rely on numerous classical image features that lead to high computational costs given the very large resolution of these images. This paper presents a detection method with very low computational complexity that is able to identify the skin depth at which the lentigo can be detected. The proposed method performs multiresolution decomposition of the image obtained at each skin depth. The distribution of image pixels at a given depth can be approximated accurately by a generalized Gaussian distribution whose parameters depend on the decomposition scale, resulting in a very-low-dimension parameter space. SVM classifiers are then investigated to classify the scale parameter of this distribution allowing real-time detection of lentigo. The method is applied to 45 healthy and lentigo patients from a clinical study, where sensitivity of 81.4% and specificity of 83.3% are achieved. Our results show that lentigo is identifiable at depths between 50μm and 60μm, corresponding to the average location of the the dermoepidermal junction. This result is in agreement with the clinical practices that characterize the lentigo by assessing the disorganization of the dermoepidermal junction.

This paper deals with the characterization and the classification of reflectance confocal microscopy images of human skin. The aim is to identify and characterize the skin lentigo, a phenomenon that originates at the dermo-epidermic junction. High resolution images are acquired at different depths of the skin. In this paper, an analysis of confocal images is performed for each depth and the histogram of pixel intensities in the image is determined. It is modeled by a generalized gamma distribution parameterized by a translation, scale and shape parameters ( , and ). These parameters are estimated using the natural gradient descent algorithm and used to achieve the classification between healthy and lentigo patients of clinical images. The obtained results show that the scale and shape parameters (beta and rho) are good features to identify and characterize the presence of lentigo in skin tissues.

This paper presents a fast approach for penalized least squares (LS) regression problems using a 2D Gaussian Markov random field (GMRF) prior. More precisely, the computationoftheproximityoperatoroftheLScriterionregularizedby different GMRF potentials is formulated as solving a Sylvesterlike matrix equation. By exploiting the structural properties of GMRFs, this matrix equation is solved column-wise in an analytical way. The proposed algorithm can be embedded into a wide range of proximal algorithms to solve LS regression problems including a convex penalty. Experiments carried out in the case of a constrained LS regression problem arising in a multichannel image processing application, provide evidence that an alternating direction method of multipliers performs quite efficiently in this context.

Satellite networks are easy-to-deploy solutions to connect rural un-served and underserved areas. But satellite latency has a significant negative impact on performance. Hybrid networks, combining high-throughput long-delay links (e.g. GEO satellites) and short-delay low-throughput links (e.g. poor ADSL), can improve user experience by the use of intelligent routing. Emerging solutions, such as MultiPath TCP (MPTCP), already optimize the throughput in these hybrid networks. However, this kind of solutions does not take into account QoE requirements by the lack of relevant flows information, leading to sub-optimal path selection. This paper proposes an architecture able to retrieve the content size through interconnection with Content Delivery Networks (CDNs). Then, we conduct an analytical study of a probabilistic and a size threshold based routing schemes with the Mean Value Analysis (MVA) method. This shows the great benefit brought by size information in terms of QoE. To solve the limitations due to the threshold configuration, we propose a third algorithm that takes into account the path delay and capacity. Finally, we develop a testbed in order to validate our model and to compare this third scheme to the previous ones. We obtain results equivalent to the size threshold scheme, without its disadvantages.

We are facing a steady increase in Internet usages and bandwidth requirements. However, many terrestrial infras-tructures remain unchanged due to exorbitant modernization costs, particularly in rural areas. In front of the aging of their infrastructures, concerned users are turning towards satellite internet access. Indeed, these solutions offer a high-throughput internet access at a moderate cost of deployment. Unfortunately, the long delay introduced by GEO satellite degrades significantly the user Quality of Experience (QoE) in many cases. In this paper, we consider a hybrid access, composed of a low data rate terrestrial path and a satellite path, within a Content Delivery Network (CDN) context. Thanks to the CDN, requested content size is known allowing to use this information in routing schemes to maximize the users' QoE. Then, we compare flow and packet based routing in this heterogeneous and size-aware context. We finally conclude on the limited interest in packet-based routing compared to flow-based routing, privileged by its simplicity and good performance from a QoE point of view.

Les produits d'intermodulation créés par les non-linéarités actives et passives ont des effets de plus en plus importants dans les charges utiles de satellites de télécommunications et dans les stations de base de téléphonie mobile à cause de l'augmentation du nombre de bandes utilisées, de la largeur de bande et de la puissance émise. Ils peuvent dans certains cas perturber gravement le fonctionnement des récepteurs. La modélisation de non-linéarités héritée de celle des amplificateurs donne de très mauvais résultats qualitativement et quantitativement, ce qui amène à des sur-spécifications d'équipements. Les modèles comportementaux non-analytiques (présentant une discontinuité en 0) proposés en 2010 donnent par contre de très bons résultats mais leur explication physique n'existait pas. La modélisation physique commence à donner des résultats que l'on peut comparer aux résultats de mesure et aux résultats des modèles comportementaux. La modélisation des non-linéarités des isolateurs et des circulateurs par le phénomène d'hystérésis magnétique rentre dans ce cas. Le modèle physique obtenu alors est non-analytique comme les bons modèles comportementaux. Il est correct qualitativement mais reste moins bon quantitativement qu'un modèle comportemental. Les modèles comportementaux prédisent mieux les résultats de mesure, on peut les rapprocher des modèles physiques mais ils nécessitent alors des modifications empiriques de certains paramètres cruciaux des modèles physiques non-analytiques. La compréhension de l'origine physique de ces différences permettra de proposer des modèles physiques plus proches de la réalité.

Face à la croissance du nombre d'utilisateurs par satellite, il devient nécessaire d'augmenter la bande passante disponible sur une région. Les systèmes satellites multi-faisceaux avec une répartition des fréquence en 4 couleurs ont déjà permis d'énormes progrès, mais limite la portion de la bande disponible dans chaque spot. Dans nos travaux, nous envisageons une utilisation plus agressive de la bande passante. L'interférence alors générée étant trop importante, nous nous sommes intéressés aux différentes techniques de coordination d'interférence. Pour cela, nous avons abordé plusieurs approches, d'une analyse des schémas de colorations à la formulation et résolution de problèmes d'optimisation permettant d'approcher la borne supérieure de la capacité d'un système.

Cooperative systems used for vessel identification and localization in the context of maritime surveillance, such as the Automatic Identification System (AIS), are often coupled to systems that allow the observation of uncooperative ships such as the Synthetic Aperture Radar (SAR). The combination of information coming from the SAR image and AIS signals can improve the detection of some ships in dense environments, but also allows possible piracy scenarios to be identified. The most common approach for data fusion is the “fusion after detection”, where each system processes the raw data independently. In the context of AIS and Radar, three levels of fusion can be identified: 1) fusion of the raw data, 2) fusion of raw data from a system with the processed data (list of detection) from the other system, 3) fusion of the detection lists formed by the two systems. We will focus on the first two cases, since the last case has been more widely covered in the literature. After introducing the AIS and Radar systems for maritime surveillance, we present structure of AIS data and radar signals, as well as the signal processing used to decode these AIS signals or to produce a radar image. The second chapter presents the potential benefits of the joint use of raw data from both radar and AIS for ship detection. After having described the signal models associated with the unknown ship position, we investigate the detection problem using a Generalized Likelihood Ratio Test (GLRT). The theoretical performances of this test are evaluated and allow us to estimate the performance gain in comparison to a single RSO processing. These theoretical results are validated by Monte Carlo simulations using Receiver Operational Characteristics (ROC). The detection results obtained using the GLRT are encouraging. However, the time implementation of these methods for practical applications is complicated. We therefore proceed to a sub-optimal detector using raw data from the radar and a list of detections from the AIS system, leading to a more simple detection strategy. The third chapter studies the fusion of raw radar data with a list of ship positions, formerly provided by the AIS system. Since the ships are moving and the AIS and Radar measurements are not are not acquired at the same time instants, the ship positions have to be extrapolated. Two extrapolation cases are considered in this work: 1) extrapolation errors are lower than the radar resolution and do not have to be integrated in the model, 2) extrapolation errors are not negligible and have to be taken into account in the model. Contrary to the second chapter, four hypotheses can now be considered. Indeed, in addition to the classical detection scenarios by both systems, we can identify the cases where only one of the systems detects a ship, which corresponds to the situations where a ship does not transmit its AIS position or where a ship intentionally false its AIS position. The problem can then be formalized with two successive binary hypothesis tests. This process allows the information coming from AIS and radar data to be fused naturally, aleading to improved radar detection performance. A performance comparison of this detector that uses a priori information with conventional radar detection shows that it is less sensitive to the proximity to other ships and to the ship density of the considered scenario. The fourth chapter presents the signal simulator considered in this thesis to test the detection algorithms in different surveillance scenarios, i.e., a piracy ship hijacking scenario, an illegal cargo transshipment and a navigation in a dense environment.