Cooperative Game-Theoretic Optimization for Simultaneous Structure Preservation and Noise Suppression in Image Enhancement

Low-light images typically suffer from reduced visibility, blurred edges, and corrupted details. Traditional Retinex-based enhancement frameworks have been widely developed, yet they expose a fundamental trade-off: structure-preserving enhancement often amplifies noise, while noise suppression tends to oversmooth critical structures. Addressing this antagonism requires a principled framework that can balance both objectives within a unified formulation. In this study, we develop a cooperative game approach for image enhancement, where structure preservation and noise suppression are formulated as cooperative players in a game. The payoff function jointly incorporates sparse-gradient structural priors and statistical noise characteristics, and the enhancement result is obtained as a Nash equilibrium solution of the game. Concretely, we introduce a sparsegradient- guided adaptive weighting strategy for illumination estimation, and integrate a Plug-and-Play denoising prior into reflectance restoration. The two tasks are iteratively optimized through an alternating scheme, which is theoretically proven to converge under mild conditions. Comprehensive experiments on standard benchmark datasets demonstrate that our method achieves a superior balance between structural fidelity and noise suppression, consistently outperforming state-of-the-art Retinex-based methods in both quantitative metrics and visual quality. Beyond empirical performance, this work highlights the potential of game-theoretic modeling in addressing coupled tasks in image restoration problems.