Color–polarization filter array image demosaicing: linear minimum mean squared error augmented by anisotropic diffusion

Linear minimum mean square error can be used to demosaic images from a color–polarization filter array (CPFA) sensor. Despite its good performance, the reconstruction produces high-frequency artifacts. An additional refinement step could enable enhancement of both the quantitative and visual quality of the demosaiced image. We propose a complete demosaicing framework by first studying the model selection for linear minimum mean square error using cross-validation techniques and then optimizing the anisotropic diffusion parameters. The results show that the training model converges quickly and that the refinement step enables the reduction of the edge artifacts. We also demonstrate that the proposed demosaicing method performs better compared with a dedicated CPFA demosaicing algorithm in terms of peak signal-to-noise ratio.