1: \begin{abstract}

2: We study the problem of deconvolution for light-sheet

3: microscopy, where the data is corrupted by spatially

4: varying blur and a combination of Poisson and Gaussian

5: noise. The spatial variation of the point spread function

6: (PSF) of a light-sheet microscope is determined by the

7: interaction between the excitation sheet and the detection

8: objective PSF. First, we introduce a model of the image

9: formation process that incorporates this interaction,

10: therefore capturing the main characteristics of this imaging

11: modality. Then, we formulate a variational model that

12: accounts for the combination of Poisson and Gaussian noise

13: through a data fidelity term consisting of the infimal

14: convolution of the single noise fidelities, first introduced

15: in~\fullcite{calatroni2017infimal}. We establish convergence

16: rates in a Bregman distance under a source condition for the

17: infimal convolution fidelity and a discrepancy principle for

18: choosing the value of the regularisation parameter. The

19: inverse problem is solved by applying the primal-dual hybrid

20: gradient (PDHG) algorithm in a novel way. Finally, numerical

21: experiments performed on both simulated and real data show

22: superior reconstruction results in comparison with other

23: methods.

24: \end{abstract}

25: