Improved hybrid demosaicing and color super-resolution implementation using quasi-Newton algorithms

Super-resolution algorithms can be used to reconstruct a high-resolution high-quality image from a set of low-quality images. A novel hybrid demosaicing and color super-resolution approach proposed by Farsiu, Elad, and Milanfar relies on the minimization of a nonconvex multiterm objective function using a rudimentary fixed step-size steepest-descent approach. In this paper, we show that improved performance can be achieved by implementing this approach in terms of powerful quasi-Newton algorithms.