A Demosaicing Algorithm Based on Local Directional Gradients for Polarization Image

A demosaicing algorithm based on local directional gradients for polarization image is presented. We use the pseudo-panchromatic image, which is calculated through the raw polarization image. During estimating the pseudo-panchromatic image, the directional gradient is used to preserve the details of it. Next, the interpolation algorithm based on the local directional gradient is performed to predict the missing values. Compared with the latest related methods, the proposed method reduces the root mean squared error by 62% and improves the structural similarity by 2%. Subjectively, we find that the degree and angle of linear polarization image produced by the proposed method have smaller errors and higher clarity.