Technical perspectiveComputational photography on large collections of images

THIS PAPER WILL strike a familiar chord with anyone who has ever taken a picture. The problem is easy to under-stand—replacing unwanted parts of a photograph. The authors start with an interesting twist, by making a distinction between data that " should have been there " but was obscured by a telephone pole, and data that " could have been there, " meaning that it constitutes an incorrect but plausible picture. This is a difficult problem, because faked pictures are relatively easy to spot, as recent scandals in photo-journalism have proven. Nevertheless, Hays and Efros obtain impressive results ; check out Figure 1 online (http://