After a good deal of searching I have found a freeware which will do blind deconvolution on normal, terrestrial photographs. Blind deconvolution is commonly used in astronomical imaging to help compensate for a variety of problems in that kind of photography.
http://www.hamangia.freeserve.co.uk/Here is an example of what blind deconvolution can do for our normal pictures, this is a somewhat out of focus image of a tufted titmouse on our feeder, the second version has been processed with Unshake, a blind deconvolution freeware.
A few notes from the instructions for Unshake..
First, note that Unshake works with pictures with widths and heights of 64, 128, 256, 512... pixels. If you give it a picture with a width of 257 pixels, it will pad it out (with an averaged colour) to 512 pixels, and will take much longer to process it than it would a picture with a width of 256 pixels. (This is a property of the Fast Fourier Transform algorithm which it uses.) 2048 by 2048 is the largest image which can be processed at the moment.
Crop all borders and edges from the image, and try to avoid features added after the picture was taken, such as writing. The reason for this is that such features are usually sharp, and so become distorted when the rest of the image is sharpened.
Try not to adjust the brightness and contrast of the picture, or if you must, ensure that the "gamma" is set to be linear - this may be indicated by a straight line on a graph. Failure to do this puts ripples or ghosts around edges with high contrast.
A disclaimer: Unshake has difficulty with images which are overexposed, underexposed, twisted (meaning that the image turned round an axis between the camera and the subject), or covered in fluff. So don't try to process images of playful black kittens in coal-sheds at night, after the kitten has walked across your scanner. It took me some experimentation to get good results from Unshake, it's not particularly intuitive or user friendly and larger images can take considerable time to process, there is a *lot* of math going on here and that that takes processor power.
Another example, this time of getting finer detail out of an already fairly decent image..