It is typically done by determining the 3D positions of individual imaged beads and calculating the average of fluorescent intensities in 3D. The PSF Generator you mentioned calculates a theoretical PSF depending on the chosen model for wide-field microscopy. The group that develops the PSF Generator has published a paper "3-D PSF Fitting for Fluorescence Microscopy: Implementation and Localization Application" which can be helpful for you.

Thank you Lei Xu I needed some more help with psf and have sent you a message for the same.

Arti Tyagi, instead of replying to you privately, I write my reply here so that it is open to the public in case it can be helpful for others as well. For theoretical calculation, you can download PSFGenerator.jar as a standalone application from http://bigwww.epfl.ch/algorithms/psfgenerator/ and use it directly to generate a PSF. I personally recommend Gibson & Lanni PSF model as it models a typical experimental wide-field configuration without too many complicated parameters. The website also provides a very good guide for setting each parameter. It should be straightforward to use it.

To determine the PSF experimentally, you simply have to take a 3D stack of acquired images of many fluorescent beads, crop each individual bead image out with the same size while ensuring the peak intensity for each bead image is centered, and then average the bead images to get your experimental PSF.

With a PSF at hand, determined either from theoretical calculation or from fluorescence images of beads, you typically perform a deconvolution of fluorescence images of your biological specimens using the PSF. The same group which develops 'PSFGenerator' also publishes a java program to do deconvolution 'DeconvolutionLab2' : http://bigwww.epfl.ch/deconvolution/deconvolutionlab2/ .

As in their paper 'DeconvolutionLab2 : An Open-Source Software for Deconvolution Microscopy', they have included many different algorithms for deconvolution. You need to make a choice of which algorithm to use. Read the instructions on their website about how to use the software. By the way, Richardson–Lucy algorithm is very commonly used for deconvolution. In case you find it difficult to make your choice for deconvolution algorithms, you can always use this one.

If you have questions for more details regarding how to use these softwares, it is perhaps easier for you to simply ask the original authors.