To compare the different restoration algorithms on a Carl Zeiss Vision KS400 system, an f-actin sample 512x512x30 (courtesy of Dr. Kem Rogers, University of Western Ontario) was imaged. The sampling distances where 0.11mm (transverse) and 0.4 mm (axial), resulting in a total volume of approximately 56x56x12 mm. For the acquisition a Zeiss LSM 410 was used with a Zeiss plan apochromat 63x / 1.4. The image was scanned once with the pinhole wide open and for comparison reasons with the pinhole closed to a size of approximately 1 Airy unit. The acquisition took place under standard working conditions with the intensity scaling slightly incorrect so that one smaller region became saturated. A measured PSF was not available, so the internally calculated theoretical one (vector wave theory) was used. The representation of the images here is with a maximum intensity projection and the meridional views are scaled to match the transverse scaling (isotropic conditions).

 Raw images
Left: non-confocal (similar to widefield) [real size]; Right: confocal, pinhole ~1 AU [real size]

 Restored non-confocal image using the iterative constrained Maximum Likelihood algorithm
Poisson Likelihood / Tikhonov regularization, 20 iterations conjugate gradient;
Left: scaled linear 0-255 [real size]; Right: scaled logarithmically to show finer detail in lower intensities [real size].

 Restored non-confocal image using the iterative constrained Least Squares algorithm
Gaussian Likelihood / Tikhonov regularization, 20 iterations conjugate gradient;
Left: scaled linear 0-255 [real size]; Right: scaled logarithmically to show finer detail in lower intensities [real size].

 Restored non-confocal image using the iterative constrained Jansson - van Cittert algorithm
20 iterations, intensity scaled linear 0-255 [real size].

 Restored non-confocal image using a regularized inverse Filter
intensity scaled linear 0-255 [real size].

 Restored non-confocal image using Nearest Neighbour algorithm
intensity scaled linear 0-255 [real size].

With respect to subjectively perceived image quality, the maximum likelihood algorithm seems to provide the best result, which shows to become comparable to the confocal raw data.