Color demosaicking with an image formation model and adaptive PCA

Color demosaicking is an ill-posed inverse problem of image restoration. The performance of a color demosaicking algorithm depends on how thoroughly it can exploit domain knowledge to confine the solution space for the underlying true color image. We propose an ℓ1ℓ1 minimization technique for color demosaicking that exploits spectral and spatial sparse representations of natural images jointly. The spectral sparse representation is derived from a physical image formation model; the spatial sparse representation is based on a windowed adaptive principal component analysis. In some of most challenging cases of color demosaicking, the new technique outperforms many existing techniques by a large margin in PSNR and achieves higher visual quality.

► We propose a novel demosaicking method with image formation model and adaptive PCA. ► The spectral sparse representation is derived from a physical image formation model. ► The spatial sparse representation is based on a windowed adaptive PCA.