How signal processing can improve the quality of division of focal plane polarimetric imagers

Polarimetric imaging can be done with a division of focal plane (DoFP) camera. This type of camera uses a grid of superpixels. Each superpixel consists of four neighbor pixels with four polarizers having different orientations in front of them. Thus, this kind of camera enables to estimate the linear Stokes vector in a single acquisition. Full stokes polarimetric imaging can be realized by adding a retarder in front of the DoFP camera and performing at least two acquisitions with two different values of retarder orientation. The effective retardance of the retarder depends on several parameters such as temperature and wavelength, which are not always controlled when using such a camera on the field. Therefore, this retardance may not be known precisely, and using a retardance value different from the true one will lead to a bias in estimating the Stokes parameter S3, which contains the information about circular polarization. This bias may become greater than the estimation standard deviation due to noise and thus have a significant impact on estimation. We demonstrate that thanks to measurement redundancy, it is possible to calibrate this retardance directly from the measurements, provided that three acquisitions instead of two are performed and the signal to noise ratio is sufficient. This autocalibration totally cancels the bias and yields a Stokes vector estimation variance identical to that obtained with the true value of the retardance. We study the practical conditions under which this method can be applied, perform experimental validation of its performance, and propose a criterion to decide if it can be applied depending on the acquired measurements.