The basic process used is 'Deconvolution" and well-known. You can get a quick synopsis from wikipedia. A more thorough explanation of this process can be found in almost any signal or image processing text.

Basically, if you try and take an image of a very small point (like a very narrow beam laser) of know characteristics, that image is spread or blurred as it passes thru the lens onto the sensor (or film). The blurring is referred as the point spread function (PSF). So you have the original image multiplied by the PSF (convolved in the frequency domain) to get the final blurred image.

At NASA (or Langley), the camera/lens systems are physically tested to get a very accurate measure of it's PSF. The better the PSF (and noise) is known, the better the original image can be recovered from the recorded image. The process is to divide the recorded image by the PSF (deconvolved in the frequency domain) to recover the original image. The recovered image will never be the same as the original unless you have a perfect measurement of the PSF and no noise. But it will be an improvement.

So your question is 'why hasn't it been incorporated into other products today?' Well it has, sort of.

The problem with incorporating this process into general software is that the PSF and noise is unknown for any given recorded image. These various software applications that purport to use deconvolution rely on what's called "blind deconvolution", even though they don't advertise this. They use basic estimates of an unknown PSF, either fixed or derived iteratively, to deconvolve with the recorded image. Gaussian-like functions are used with varying spreads to estimate a PSF. Some just refer to those as different model selections. Blind Deconvolution has not been highly effective as a general image restoration method for a number of reasons. The PSF of camera systems are complex with a number of lens characteristics contributing to image blurring, sensor characteristics and unknown noise characterization. Improvements in the deconvolution algorithms, such as Weiner deconvolution, Richardson-Lucy deconvolution, Van Cittert deconvolution, etc., has improved the process but is mainly used in more complex, specialized software.

The concept is well known, but good implementation is difficult requiring a good measure of the PSF and is computationally intensive since it's usually done in the frequency domain. Blind Deconvolution in general photography provides very limited benefits and creates very messy results when overdone.

Referring back to the OP, if simple sharpening can't fix it, the delete key is your best friend.

Glad to help. Been using these techniques, primarily in the field of spectroscopy, for the last 35 years prior to retiring. I know this subject deviated somewhat from the OP's original post, but what the heck; it's still interesting.

Another misconception I read in this thread was that the information in a blurred image is lost, which it isn't. As it passes thru an optical system, it's just rearranged by the true PSF. Using a good estimate of the PSF allows the re-rearrangement back to close to the original image (i.e. linear systems are reversable). Used extensively in the scientific community, especially in astronomy, microscopy, medicine and forensics to name a few.

Happy new year JD750, wish you and your family all the best in 2023.