We propose a novel acquisition scheme and non-parametric multi-image based method for correcting illumination in fluorescence images. Our approach measures changes in intensity of the subject by moving the microscope stage at regularly spaced intervals, and exploits this information to learn the correction function. The acquisition process and learning are performed prior to imaging, and take only a few minutes. Afterwards, images can be corrected for vignetting and photobleaching effects on the fly. Our approach can be implemented in any microscope with a motorized stage, and does not require a reference calibration slide. Experiments demonstrate that our method outperforms standard approaches to illumination correction.