In the last decades, fluorescence microscopy has evolved into a powerful tool for modern cell biology and immunology. However, while modern fluorescence microscopes allow to study processes at subcellular level, the informative content of the recorded images is frequently constrained by the limited dynamic range of the camera mounted to the optical system. In addition, the quality of acquired images is generally affected by the typically low-light conditions that lead to comparatively high levels of noise in the data. Addressing these issues, we introduce a variational method for high dynamic range (HDR) imaging in the context of fluorescence microscopy that explicitly accounts for the Poisson statistics of the unavoidable signal-dependent photon shot noise and complements HDR image reconstruction with edge-preserving denoising. Since the proposed model contains a weight function to confine the influence of under- and overexposed pixels on the result, we briefly discuss the choice of this function. We evaluate our approach by showing HDR results for real fluorescence microscopy exposure sequences acquired with the recently developed MACSima^TMSystem for fully automated cyclic immunofluorescence imaging. These results are obtained using a first-order primal-dual implementation. On top of this, we also provide the corresponding saddle-point and dual formulations of the problem.

在过去的几十年里，荧光显微镜已经发展成为现代细胞生物学和免疫学的强大工具。然而，虽然现代荧光显微镜允许在亚细胞水平上研究过程，但记录图像的信息内容经常受到安装在光学系统上的相机的有限动态范围的限制。此外，采集图像的质量通常会受到通常的低光照条件的影响，这会导致数据中的噪声水平相对较高。针对这些问题，我们在荧光显微镜的背景下引入了一种用于高动态范围 (HDR) 成像的变分方法，该方法明确解释了不可避免的依赖于信号的光子散粒噪声的泊松统计，并用边缘保留去噪补充了 HDR 图像重建。由于所提出的模型包含一个权重函数来限制曝光不足和过度曝光像素对结果的影响，我们简要讨论该函数的选择。我们通过显示使用最近开发的 MACSima 获得的真实荧光显微镜曝光序列的 HDR 结果来评估我们的方法用于全自动循环免疫荧光成像的^TM系统。这些结果是使用一阶原始对偶实现获得的。除此之外，我们还提供了问题的相应鞍点和对偶公式。