Color demosaicking with the spatial alignment property of spectral Laplacians

Abstract

We present a color demosaicking method which aims to improve the demosaicking performance by thoroughly exploiting domain knowledge to confine the solution space for the underlying true color image. We use a physically-induced model to infer spatial alignment property of spectral Laplacians. Specifically, the 2D Laplacians of different channels are sparse, and it benefits spatial alignment property across different color channels. Then, we establish an optimization model, using the image formation model and the spatial alignment property of spectral Laplacians, to solve the ill-posed problem of the color demosaicking. Finally, some extensive experiments have been done and show the availability of our approach in color image demosaicking.