Quadratic Approximation of Taylor Series for DoFP Polarization Image Demosaicking

Polarization demosaicking algorithms are often employed to reduce IFOV errors and recover missing polarization information with less artifacts. In this paper, we propose a demosaicking algorithm for polarization images based on the quadratic approximation of Taylor series. By considering the correlation of four channels in the polarization image, the quadratic approximation of Taylor series is utilized to estimate the missing pixel values of high correlated channels. For low correlated channels, we employ local average on it. The experiment results show the comparisons between our method and some latest methods. Our method reduces the root mean square error by 34.93% in average, meanwhile, the structural similarity index measure is improved by 9.2%. By comparing the DoLP and AoLP images, the proposed algorithm shows less noisy and artifacts than others.