Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
We investigate the problem of automatic tuning of a deconvolution algorithm for three-dimensional (3D) fluorescence microscopy; specifically, the selection of the regularization parameter lambda. For this, we consider a realistic noise model for data obtained from a CCD detector: Poisson photon-counting noise plus Gaussian read-out noise. Based on this model, we develop a new risk measure which unbiasedly estimates the original mean-squared-error of the deconvolved signal estimate. We then show how to use this risk estimate to optimize the regularization parameter for Tikhonov-type deconvolution algorithms. We present experimental results on simulated data and numerically demonstrate the validity of the proposed risk measure. We also present results for real 3D microscopy data.
, , , ,