Automized polarization imagery – autopol

The objective of the present project is to develop this type of system. The demonstrator will produce polarimetric images at 1.55 µm thanks to laser illumination. It will integrate different control strategies involving image segmentation algorithms. To address this challenge, the project team possesses skills in imaging system design, implementation and image processing.

The built demonstrator with be the first in the World to implement jointly image processing algorithm and adaptive polarization imager. Applications will be military (outdoor scene analysis, decamouflaging, object detection) and civilian (biomedical imaging, robotics or industrial inspection). At the end of the project, a prospective study will evaluate these perspectives.

* Design of the infrared imager
* Implementation of the imager
* Design of segmentation algorithms

* Implement control strategies on infrared imager
* Valider l'imageur sur des scénarios réalistes

* Présentation to Journées d'imagerie optique non-conventionnelle,
19-20 mars 2013
« Automatic polarimetric imaging system for contrast optimization using non-parametric statistical snake »
G. Anna (a), N. Bertaux (b), F. Galland (b), François Goudail (a),Daniel Dolfi (c)
(a) Laboratoire Charles Fabry, UMR 8501, Institut d’Optique, CNRS, Univ Paris Sud 11, Palaiseau
(b) Institut Fresnel CNRS, Aix- Marseille Université, Ecole Centrale Marseille, Marseille
(c) Thales research and Technology - France, Palaiseau

Active polarimetric imaging systems have demonstrated their efficiency in military (object detection, decamouflaging) and civilian (biomedical imaging, industrial control) contexts. Their added value is particularly significant in detection/identification applications since they reveal contrasts that do not appear in classical (panchromatic or spectral) images.

An active polarimetric imager illuminates the scene with light having a given state of polarization and analyzes the state of polarization of the light scattered by the scene. In most current systems, the number of available states is limited, and they cannot be modified. However, recent research has shown that it is possible to significantly increase the contrast by adapting the polarization states to the observed scene. These optimization techniques still assume that the polarimetric properties of the objects are known beforehand, which sets a limit to their practical application.

In order to fully exploit the potential of these techniques, it is thus necessary to design systems that automatically adapt to the observed scene. If the characteristics of the scene are unknown, the necessary information must come from image analysis algorithms. In order to optimize the performance of polarimetric imagers, the essential condition is thus to integrate, in a global system, fast and powerful image segmentation algorithms and an agile imaging system that makes it possible to generate any polarization state in illumination and analysis.

The objective of the present project is to develop such a system. The demonstrator will produce polarimetric images thanks to laser illumination at 1.55 µm. It will integrate different strategies (automatic or semi-supervised) for controlling the instrument parameters (illumination and analysis polarization states) with image segmentation results. To meet this challenge, the consortium is composed of one academic and one industrial team specialized in theoretical and experimental aspects of polarimetric imaging, and of one team specialized in image segmentation algorithms and their application to non-conventional images.

The demonstrator built during this project will be the first polarimetric imager to jointly implement image segmentation algorithms and a polarization-agile image acquisition system. It will be able to solve difficult outdoor scene analysis problems such as decamouflaging, obstacle or suspicious object detection. On a more general point of view, this project will make it possible to develop and validate an innovative methodology for controlling polarization states with image segmentation results. Applications of this innovation will reach far beyond military applications, to such domains as biomedical imaging, robotics or industrial control. At the end of the project, a prospective study will evaluate these perspectives.