AbstractRecent studies have shown that joint denoising and super-resolution (JDSR) approach is capable of producing high-quality medical images. The training process requires noise-free ground truth or multiple noisy captures. However, these extra training datasets are often unavailable in fluorescence microscopy. This paper presents a new weakly-supervised method, in which different from other approaches, the JDSR model is trained with a single noisy capture alone. We further introduce a novel training framework to approximate the supervised JDSR approach. In this paper, we present both theoretical explanation and experimental analysis for our method validation. The proposed method can achieve an approximation accuracy of \(98.11\%\) compared to the supervised approach. The source code is available at https://github.com/colinsctsang/weakly_supervised_JDSR.KeywordsSuper-resolutionDenoisingWeakly-supervised