Photonics.com has an article on two patent applications related to a technique to identify which digital camera took a particular photograph. The inventors are researchers at State University of New York at Binghamton (NY, USA). The lead researcher is Jessica Fridrich.
The general idea of identifying digital cameras by their 'dead' (i.e. non-functioning) pixels or other characteristics of their image sensor has been discussed over the past few years and some techniques have emerged to identify brands of cameras (due to their manufacturers' standard implementation of internal image processing and such) and even individual cameras.
The new technique described in the article is based on extracting a very weak noise pattern that is present in all digital photographic images due to inhomogeneities in solid-state sensors used. The inventors claim that their technique can extract this pattern from a single image. Since it is based on a pattern derived from the several million pixels present in most current digital cameras, they intuit that it may indeed be unique. This is a claim that has to be made very, very carefully. With scientific (in this case, forensic) analysis that may be used in court, it is especially important to not claim that a signature is "unique" until it can be proven to an accepted legal standard, such as Daubert in the USA.