Abstract Image noise is a fundamental problem in fluorescence microscopy analysis, especially in live cell imaging applications where the number of detected photons is limited due to low power of excitation lasers to prevent phototoxicity during extended imaging experiments. The noise increases measurement uncertainty and complicates further image processing routines such as deconvolution, object detection and segmentation. State‐of‐the‐art denoisers are computationally expensive and require training using large datasets, which are not available in cases of typical biological imaging experiments with rather scarce and unlabelled data. Here, we show that a denoiser can be trained using a single image containing a cropped out object of interest, where we exploit the symmetry often present in biological structures at molecular scales. As only a single example is used during training, our method can be trained even with limited computational resources, obtaining competitive denoising performance compared to the state‐of‐the‐art methods.
Wolf et al. (Fri,) studied this question.