The 3rd image in the original article compares the original and the deconvolved result. The main idea is you want to invert a linear transformation (a matrix)

The 3rd image in the original article compares the original and the deconvolved result. The main idea is you want to invert a linear transformation (a matrix) which is given by the operation of your imaging system, i.e. a blurring operation. This linear transform often has a poor condition number, however - this means that you can't really invert it, or if you try to you end up with certain components of your signal that need huge amplification, which makes noise sensitivity a major issue. One solution to this is to introduce a "prior" on your signal that represents your expectations about what the signal looks like, i.e. is it smooth, or does it have a few edges etc. Then you can better tolerate the inherent sensitivities. I'm not sure what priors they're using for this situation or if they're using any at all, so in the absence of such a description I'm taking the results with a grain of salt. See for example: http://people.seas.harvard.edu/~schan/deconvtv_folder/deconvtv_image.html http://people.seas.harvard.edu/~schan/deconvtv_folder/deconv...