High-fidelity image decolorization with pure odd polynomial subspace and joint covariance optimization
TL;DR
This paper proposes a high-fidelity decolorization algorithm based on a pure odd-degree polynomial subspace and joint covariance optimization that outperforms existing state-of-the-art approaches in both qualitative visual quality and quantitative metrics.
Abstract
Image decolorization is a fundamental task in computer vision and computer graphics, aiming to convert color images into grayscale while maximizing the preservation of visual structures and contrast. Traditional methods based on even-degree polynomials suffer from the pervasive “polarity folding” problem, which often leads to severe contrast loss in isoluminant regions. To address this issue, this paper proposes a high-fidelity decolorization algorithm based on a pure odd-degree polynomial subspace and joint covariance optimization. Specifically, we construct a seven-dimensional odd feature space to fundamentally eliminate the polarity folding effect and design a joint optimization framework that fuses global pixel distribution with local gradient information. Extensive experiments on the Čadík dataset, complex scene decolorization dataset, and Color250 benchmarks demonstrate that our method outperforms existing state-of-the-art approaches in both qualitative visual quality and quantitative metrics—including color contrast preserving ratio, color content fidelity ratio (CCFR), and E-score—providing robust and artifact-free decolorization results.