Single molecule localization microscopy code based on a deconvolution algorithm with a L0 regularization term to promote sparsity. The L0 penalized least-squares criterion is continuously relaxed with the continuous exact l0 (CEL0) functional, allowing thus its minimization using an iteratively reweighted L1 method. More details can be found in the following paper:

High density molecule localization for super-resolution microscopy using CEL0 based sparse approximation. Proc. ISBI, 2017. Simon Gazagnes, Emmanuel Soubies and Laure Blanc-Féraud.

• main function SMLMCEL0.m
• function ComputeNorm_ai.m which computes the norm of the columns of the used operator
• folder ToyExample containing a simple example of use in script.m

The algorithm has been tested on the 2D high density datasets of the SMLM challenge 2016. For these tests, algorithm parameters have been set as follows:

• coefEch: 4 (i.e. each pixel of data images is divided in 4)
• itmaxIRL: 200 (max number of iterations for outer loop IRL1)
• itmaxFista: 200 (max number iterations for inner loop FISTA)
• Gaussian PSF with variance: 4.5e-3 (for a normalized image domain [-1 1]^2)
• lambda: 1.1 (for dataset ER2.N3.HD-2D) and 0.21 (for dataset MT4.N2.HD-2D)