Physics-Inspired Single Image Detail Enhancement Based on Kirchhoff's Current Law

Residual learning-based single image detail enhancement methods often suffer from premature convergence to local optimum due to greedy search strategy. In circuit systems, current preferentially flows through the branch with the minimum equivalent resistance. At any node, the branch currents satisfy Kirchhoff's Current Law (KCL). Inspired by KCL, this letter proposes an Image Detail Enhancement (IDE) method called KCLIDE. The proposed method models the image search and patch matching process as the dynamic behavior of current within a circuit, where patch matching errors are seen as circuit resistance. Guided by KCL, the optimization path naturally converges towards regions of minimum resistance, effectively yielding the minimal reconstruction error, achieving optimality with globally set parameters. Experimental results demonstrate that KCLIDE consistently outperforms state-of-the-art methods in both visual fidelity and quantitative metrics, notably achieving PSNR and SSIM gains of 1.39 dB and 0.0163 under ×4 enhancement factor, respectively, over the recent DFCN method on the T91 dataset.