Robust Multispectral data analysis for target detection and IDentification


MultID is a scientific project financed by the ANR and DGA as part of the ASTRID ANR projects. The aim of this project is to design, study and apply multispectral data analysis tools on the context of surveillance imaging to tackle the problem of robust target detection and identification.

The teams involved in this project have long been developing and apply blind source separation methods. In a wide range of scientific research fields - ranging from geophysics, remote sensing to astrophysics - accounting for accurate/approximate spectral priors is generally becomes paramount. Unfortunalely such kind of information is rarely - if not - well accounted for. The first part of the project will then focus on the following key points :

        1 - Design and study fully supervised source separation techniques to decompose multispectral data in large-scale spectral dictionaries.

        2  - Develop semi-blind source separation techniques for robust source separation with partly known spectral priors.

The second part of this project focuses on the applications of (semi)-blind source separation techniques to surveillance imaging. In this specific domain of imaging, the detection/identification of targets or components of interest out of a non-informative background is generally a strenuous problem due the multiple sources of imaging variabilities (e.g. illumination conditions, background variability, noise ...etc). To alleviate these issues, we propose studying the performances of blind and semi-blind source separation techniques to tackle the problem of robust target detection and identification.


The MultID project in a few words ...

Contact / Info

Last updated: 12/05/2012