Wide-field fluorescence microscopy is fundamentally limited by out-of-focus background haze, which degrades image contrast and obscures subcellular details. To overcome these limitations without the complexity of specialised hardware, we present a high-performance computational sectioning framework based on a physics-informed variational model. By modelling the observed image as an additive superposition of high-frequency in-focus signals and a low-frequency defocused background, we employ a penalised least-squares optimisation to achieve precise signal-background separation. To bridge the gap between rigorous baseline estimation theory and real-time imaging requirements, we introduce a Gaussian-equivalent regularisation strategy. This approach replaces traditional implicit penalty operators with explicit Gaussian smoothing convolutions, significantly enhancing computational efficiency while maintaining the deterministic nature of the iterative optimisation. Quantitative evaluation against confocal ground truth demonstrates that our method increases the Resolution Scale Pearson (RSP) coefficient by 1.5-fold and reduces the Resolution Scale Error (RSE) by a factor of six. Benchmarked against DeepMRA and HiLo techniques, our framework provides a robust, physically interpretable, and low-cost solution for high-contrast deep-tissue imaging.
He et al. (Mon,) studied this question.