Search results for " Computer science"

Perspective-n-Learned-Point: Pose Estimation from Relative Depth

2019

International audience; In this paper we present an online camera pose estimation method that combines Content-Based Image Retrieval (CBIR) and pose refinement based on a learned representation of the scene geometry extracted from monocular images. Our pose estimation method is two-step, we first retrieve an initial 6 Degrees of Freedom (DoF) location of an unknown-pose query by retrieving the most similar candidate in a pool of geo-referenced images. In a second time, we refine the query pose with a Perspective-n-Point (PnP) algorithm where the 3D points are obtained thanks to a generated depth map from the retrieved image candidate. We make our method fast and lightweight by using a commo…

Implémentation en temps réel d'une architecture embarquée pour le comptage de personnes dans une foule

2018

International audience

TRAITEMENT NUMÉRIQUE D'IMAGES ET APPLICATIONS: Méthodes statistiques optimisées pour le traitement numérique des images de grandes tailles,

2010

Automatic Detection of Cerebral Microbleed in SWI Using Radon Transform

2013

International audience; Since presence and number of cerebral microbleeds (CMBs) have come to attention as a potential biomarker, an automated scheme to improve visualization is required. In this work, a new approach of CMB identification in SWIs is presented and compared to visual rating. The method relies on two main steps: a 3D anisotropic multi-scale approach that extracts size and centre of all potential CMBs within the image, and feature extraction using the Radon Transform for final classification using a random forest classifier. The novelty of the technique consists in combining Radon transform and multiscale analysis to obtain robust feature descriptors.

HAZE AND CONVERGENCE MODELS: EXPERIMENTAL COMPARISON

2015

International audience; Bad environmental conditions like bad weather, such as fog and haze, and smoke-filled monitored closed areas, cause a degradation and a loss in contrast and color information in images. Unlike outdoor scenes imaged in a foggy day, an indoor artificial hazy scene can be acquired in controlled conditions, while the clear image is always available when the smoke is dispersed. This can help to investigate models of haze and evaluate dehazing algorithms. Thus, an artificial indoor scene was set up in a closed area with a mean to control the amount of haze within this scene. While a convergence model simulates correctly a small amount of haze, it fails to reproduce the sam…

A new system for watermarking based on the turbo-codes and wavelet 5/3

2012

International audience; In order to contribute to the security, transmission and sharing of medical data between hospital centers, we propose a watermarking scheme for hiding medical information, specific to the patient, in the host image. In this regard, the watermark consists of inserting data (usually binary data) in media (image in our case) and try to find these data with maximum fidelity, after applying various attacks. The principal goal of our proposal is the conservation of psychovisual quality of the image, on the one hand, and the extraction of the different data inserted in another side The robustness of the proposed scheme is tested against various attacks such as JPEG compress…

European Programs and their Extension in the field of Computer Vision, Color and Robotics

2010

International audience; Over the last 4 years, the University of Burgundy (uB) has intensified its international oriented actions in the field of Computer Vision and Robotics through further development of the educational network, partnerships, programs and communication. The creation of European programs, such as the Erasmus Mundus program and the Thematic Networks (Life Long Learning) allowed us to develop new programs, building up valuable academic and project administration experience as well as developing new actions for incoming foreign students. In this paper we describe some of our activities and those of the University of Saint Etienne (UJM) and University of Genova (UNIGE), and th…

A Watermarking of Medical Image: Method Based "LSB"

2011

International audience; In this paper, we present a new approach for watermarking of medical image that we are trying to adapt to telemedicine. This approach is intended to insert a set of data in a medical image. These data should be imperceptible and robust to various attacks. It's containing the signature of the original image, the data specific to the patient and his diagnostic. The purpose of the watermarking method is to check the integrity and preservation of the confidentiality of patient data in a network sharing. This approach is based on the use the LSB (least significant bits) of the image and tools borrowed from cryptography.

Méthode de "Multicouches" Pour le Tatouage des Images Médicales

2012

International audience; Dans le but de contribuer á la sécurité du partage, au transfert et á l'indexation des images médicales, on a mis en œuvre une nouvelle approche de tatouage. Cette approche se base sur l'utilisation de diverses techniques tels que la technique du Code Division Multiple Access (CDMA), de la Transformation en Ondelette Discrète (TOD), des Codes Correcteurs d'Erreurs (CCE) et de la fonction de hachage (MD5),... Le but de notre contribution étant d'augmenter le taux d'intégration tout en conservant l'aspect visuel de l'image d'une part et de détecter et même corriger les différentes altérations qui peuvent avoir lieu de l'autre part. Cette approche sera testé sur des ima…

A Machine Learning Approach for Computer-Aided Detection of Cerebral Microbleed Using High-order Shape Features

2014

International audience; This paper presents a novel machine learning approach for computer-aided detection of microbleeds in SWI. The major contributions are: identifying microbleed extent in order to extract proper cubic regions-of-interest (ROI) containing the structure, (2) extracting a set of robust 3- dimensional (3D) Radon- and Hessian-based shape descriptors within the ROIs as well as 2D Radon features computed on intensity-projection images of the corresponding ROIs, and (3) incorporating a cascade of random forests (RF) classifiers to iteratively reduce false detection rates while maintaining a high sensitivity.