A multi-angle polarization imager (MAPI) can obtain multi-spectral and multi-angle polarization information, which is used to retrieve the microphysical properties of aerosols and clouds. Polarization high-precision detection requires accurate polarization calibration. Since MAPI is without an on-orbit calibrator, we use natural scenes as the sources of polarization calibration. A full-field on-orbit polarization calibration model is established, and the polarization characteristics of water clouds are analyzed. Water cloud pixels with a scattering angle of 100°are selected as unpolarized calibration sources. The relative transmittance of the polarizer-filter combination, the polarization degree of the optical lens, and the low frequency relative transmittance required by the polarization response matrix are solved, and finally the polarization response matrix of the full image is calculated. The average value of relative transmittance of the polarizer-filter combination for three years is analyzed and verified by laboratory calibration results. The relative error is 0.71%. The least squares method is used to fit the variations of the polarization degree of the optical lens and the low frequency relative transmittance with the angle of view. The laboratory polarization calibration results are used to verify that the relative error of the central field of view is 1.22%. The uncertainty of the algorithm is analyzed. The synthetic uncertainty of the central field of view and the marginal field of view is 1.27% and 2.19%, respectively, which meet the design specifications. The calibration method can be used as a reference for the on-orbit calibration of wide-field polarized imaging devices, and provides quality assurance for aerosol inversion.