Abstract : The polarimetric information is fundamentally similar for the micro-wave (radar), infrared and visible (optics) domains. However, measurement instruments vary considerably and therefore all the developed tools (calibration, geometric conventions, noise filtering) differ between the different communities. These differences are illustrated in particular by the following : the acquisition geometry, typically bistatic in optics and mainly monostatic in radar ; the nature of the measure, coherent for the radar (intensity and phase) and incoherent for optics ; the processing. This thesis attempts to compare the use of these tools in these two worlds to improve our understanding and use of polarimetry. A calibration method inspired by optics was first adapted to the case of bistatic radar. Then the impact of bistatic geometries and choice of reference polarimetric bases were analyzed, first for simple targets and then for depolarizing targets. That leads us to choose the diffusion plane as reference plane for conducting polarimetric analysis on bistatic radar images. Then we focus on the estimation of the covariance matrix in radar and on the statistical pre-processing. We propose here an original polarimetric segmentation scheme and a new family of polarimetric detectors. Finally, we discuss the different polarimetric decompositions, and various applications : characterization of depolarizing optical media (biological tissues) or radar (forest), help for 3D rendering.