Color image demosaicking using inter-channel correlation and nonlocal self-similarity

• We formulate the demosaicking problem in a perspective of image reconstruction.
• We model color images by considering different types of prior information.
• The joint modeling-based minimization problem is solved efficiently.
• A block-based demosaicking framework is designed to work with JM-CDM.
• Our algorithm outperforms many state-of-the-arts demosaicking methods.

Color demosaicking is used to reconstruct full color images from incomplete color filter array samples captured by cameras with a single sensor array. In reconstructing natural-looking images, one key challenge is to model and respect the statistics of natural images. This paper presents a novel modeling strategy and an efficient color demosaicking algorithm. The approach starts with joint modeling of the color images, which supports simultaneous representation of inter-channel correlation and structural information in an image. The inter-channel correlation is explored by measuring the channel difference signals in the gradient domain, while the structural information is explored by nonlocal low-rank regularization. An efficient algorithm is then proposed to solve the joint formulation, by dividing the minimization problem into two sub-problems and solving them iteratively. The effectiveness of the proposed approach is demonstrated with extensive experiments on both noiseless and noisy datasets, with comparison with existing state-of-the-arts color demosaicking methods.