It's quite possible. Not only is it possible to perform deconvolution, but since you know that you're looking for text data, and you even have the font, you can

It's quite possible. Not only is it possible to perform deconvolution, but since you know that you're looking for text data, and you even have the font, you can do much better. You can iterate through millions of names very quickly and find which one, once blurred produces the best match. In the same vein, check out http://www.ee.columbia.edu/~wliu/CVPR05_LiuWei1.pdf http://www.ee.columbia.edu/~wliu/CVPR05_LiuWei1.pdf