Color-to-Gray Conversion with Perceptual Preservation and Dark Channel Prior

Abstract

This paper aims to present a decolorization strategy based on perceptual consistency and dark channel prior. The proposed model consists of effective fidelity terms and a prior term. We use the $\mathcal{l}$₀-norm to control the sparsity of the dark channel prior. To solve the non-convex minimization problem, we employ the split and penalty technique to simplify the minimization problem and then solve it by the carefully designed iteration scheme. Besides, we show the convergence of the algorithm using Kurdyka-Lojasiewicz property. The numerical evaluation in comparison with other state-of-the-art methods demonstrates the effectiveness of the proposed method.