基于物理引导的图像亮度增强神经网络研究OA

To address the challenges of insufficient brightness and the difficulty of simultaneously correcting noise and color deviation in low-light images,this paper proposes a physics-guided image luminance enhancement neural network.The proposed method decomposes the image into luminance and chrominance components within a linear space,and forms a physical backbone with a monotonic tone curve controlled by a few hyperparameters,while a residual network learns only amplitude-limited local compensation.Furthermore,multiple physical constraints are introduced as loss functions.Experi-mental results on a low-light street scene dataset demonstrate that the model achieves superior performance.The re-search confirms that combining interpretable physical models with data-driven models significantly improves the natural-ness and stability of the enhancement results.Meanwhile,it provides a technical foundation for cross-scenario applications.