|Topic 1: Quality and Representation of Multimedia Content|
Visual Perception Models for Visual Data Analysis and Processing
We develop perceptual models of visual information representation and extraction of descriptors and visual attributes
(contrast, visibility map and visual masking) for processing, analysis and coding of visual information. The first stages
of the humanvisual system are taken into account, in particular multi-resolution and multidirectional contrast, masking
phenomenon as well as multichannel decomposition
Visual information quality
It involves evaluating and/or improving the visual information quality (multimodal medical images, stereoscopic
medical images, HDR images, stereoscopic images, stereoscopic video, endoscopic video, etc.) either
coded/decoded, transmitted (filtering, enhancement of contrast ...). We differentiate ourselves from current trends
by avoiding the development of purely mathematical (PSNR) measures, or completely inspired by the human visual
system for which the number of parameters to be adjusted is quite large.
Coding with quality control
Source coding mechanisms are developed by integrating distortion level indicators to optimize the quality of visual
information. The degradation effects, which can result from the motion (respectively disparity) estimation stage inthe
video (respectively stereoscopic image) coding, scheme are taken into consideration.
|Topic 2: Machine Learning and Multimedia Data Mining|
Image processing and machine learning
We are interested in machine learning methods for multimedia content processing. The aim is to develop methods of
processing multimodal medical images (segmentation, denoising, super-resolution ...) based on machine learning.
These methods aim to take into account the characteristics of the human visual system through models whose
parameters are estimated by machine learning methods. We are also interested in estimating the subjective image
quality by statistical learning methods.
Data mining and social networks
The aim is to develop approaches derived from the statistical learning theory for structured data processing (numerical
attributes, texts, multimedia data) resulting from social graphs. We are interested in exploiting the knowledge of the
links between a set of actors to improve the construction of models: to plan, identify and characterize a group of actors.
|Topic 1 : R3D : Wireless Networks ( Dimensionning , Deployment, D issemination )|
Dimensionning of wirless networks
Deployment of wireless networks
|Topic 2 : RIS : Networks Infrastructure and services|
Novel paradigms for networking services