Estimation of the degree of polarization from a single image: characterization of the precision under various coherent imaging conditions

摘要

Many application fields are interested in polarimetric imagery that can reveal contrast and details invisible in standard intensity imaging. The degree of polarization (DOP) is a scalar parameter that can give interesting information about the polarimetric nature of the light, varying between 0 for an unpolarized light and 1 for a totally polarized light [1]. However, its measurement is in general quite costly and time-consuming since it requires to record four images [2,3] or two images [4,5] of the scene with different polarization devices such as polarizer and retardation plate (see Fig.1). We have shown recently that it is possible to locally estimate the DOP of the light from a single intensity image of the scene under coherent illumination (typically a laser beam) [6,7]. On the one hand, coherent light allows night vision capability and a better resolution compared to incoherent light for a given aperture size. On the other hand, the drawback lies in the speckle noise that results from interferences of the backscattered light. Nevertheless, one can take advantage of these perturbations since they contain information on the degree of polarization of the reflected light [1,6,7]. This technique can be interesting for polarization imagery applications that need the greatest simplicity and the shortest acquisition time. Indeed, after enlightenment of the scene by laser light, the proposed method only requires a single image acquisition and a very short processing time. We propose here to recall the principle of the modelization and of the estimation technique. The latter is characterized by studying its theoretical precision which is compared to the best precision expectable, derived from the determination of the Cramer-Rao bound. Eventually, some simulated experiments of estimation are presented that allow one to validate the theoretical considerations and quantify the utilization range of the proposed method.