The basic technique it uses involves two infrared cameras which determine how flat or curved each page to be scanned is and then adjusting the optical character recognition software it uses to read the text accordingly.
In other words, the infrared cameras help figure out a book’s three-dimensional shape and then back out any resulting distortions. This results in much faster book scanning since each page doesn’t need to be flattened by glass plates and spines don’t need to be broken.
The TechCrunch synopsis of the book scanning technique for which Google received patent no. 7,508,978 on 24 March of this year. The improved method, it's reasonably assumed, will improve the efficiency of preparing publications for the Google Book Project.