RGB-NIR Demosaicking Using a Two-Phase Primal-Dual Algorithm with a Laplacian Guided Image Filter Prior
Journal
2021 IEEE International Conference on Consumer Electronics-Asia, ICCE-Asia 2021
Date Issued
2021
Author(s)
Abstract
Demosaicking is a process to interpolate raw data subsampled by the color filter array into a full-resolution image. In this work, a new RGB-NIR demosaicking algorithm based on the primal-dual method is proposed. In the algorithm, we use total variance and guided image filters as the regularization terms, so that both flat and edge regions can be properly addressed. In addition, a two-phase optimization scheme is introduced to avoid problems caused by over-correlating the RGB and NIR channels. We also design an early-termination mechanism, which can stop iterations when it foresees the result could go wrong. As demonstrated in the paper, the proposed algorithm can preserve high frequency regions well while having chroma noises, halos, and artifacts effectively removed. ? 2021 IEEE.
Subjects
Laplacian guided filters
Primal-dual algorithm
RGB-NIR demosaicking
Edge detection
Infrared devices
A-Laplacian
Demosaicking
Guided filters
Guided images
Image filters
Laplacian guided filter
Laplacians
Primal-dual algorithms
Two phase
Laplace transforms
Type
conference paper