Compressed Sensing Super Resolution Of Color Images

In this work we estimate Super Resolution (SR) images from a sequence of true color Compressed Sensing (CS) observations. The red, green, blue (RGB) channels are sensed separately using a measurement matrix that can be synthesized practically. The joint optimization problem to estimate the registration parameters, and the High Resolution (HR) image is transformed into a sequence of unconstrained optimization sub-problems using the Alternate Direction Method of Multipliers (ADMM). A new, simple, and accurate, image registration procedure is proposed. The performed experiments show that the proposed method compares favorably to existing color CS reconstruction methods at unity zooming factor (P), obtaining very good performance varying P and the compression factor simultaneously. The algorithm is tested on real and synthetic images.