Credit: Carnegie Mellon Graphics
Photography and the Dark Arts? Look out Harry Potter.
James Hays and Jexei Efros are really smart people. Hays and Efros, computer scientists at Carnegie Mellon, report they have invented a whole new way of patching up pictures by "borrowing" pieces of other pictures from the web. They call the method "scene completion", but others will differ them, especially when it comes to how the "scene" gets completed -- by taking content from other pictures off the web.
Credit: Carnegie Mellon Graphics
Scene Completion Using Millions of Photographs
By using the data base of the World Web, with millions of images to pick from, Hays and Efros, have figured out that they can splice slices of reality in a seamless process that differ from previous methods.
The interesting point here is how science, which seeks to solve a problem, often complicates and creates even more problems.
As the image engineers explain:
"Our chief insight is that while the space of images is effectively infinite, the space of semantically differentiable scenes is actually not that large. For many image completion tasks we are able to find similar scenes which contain image fragments that will convincingly complete the image. Our algorithm is entirely data-driven, requiring no annotations or labeling by the user. Unlike existing image completion methods, our algorithm can generate a diverse set of image completions and we allow users to select among them. We demonstrate the superiority of our algorithm over existing image completion approaches."
To their credit, Hays and Efros, have just moved electronic photo manipulation to a whole new level -- they have given the photo industry a bigger gun in which to pass off composites, fakes, and illustrations as wondrous illusions of reality. Not that photography hasn't been dealing with these issues since its inception. It is just that this new process contributes to already growing ways in which digital shenanigans get passed off as "truthful" representations of reality. I can see the Pentagon, politicians, advertising industry, and even more conventional mainstream news operations clamoring for the software. It's all part of the slippery slope of image production in the 21th Century.
Not only are the possibilities of digital manipulation so much greater with this process, there is also the very big question as to what will constitute copyright infringement. Even if Hays and Efros use 1/1,000,000 th of a picture made by someone else, even if they borrow a few pixels here and there without asking permission or paying the owner for that 1/1,000,000th, would they be infringing on someone's copyright? What is fair use when there's a program out scanning images on the web in order to make a whole new image?
It should not come as no great surprise that science would eventually figure out a way to semantically and seamlessly reconstruct images. We already have these processes in place.
However, the implications of this new method add fuel to the already burning argument that pictures could never be trusted as faithful reflections of reality. What you get is not what was seen, but rather only a few pixels here and there of possibly millions of other images.
Thanks to Daniel Sato for the inspiration and the link.