As the worlds of artificial intelligence and digital photography collide, we’re starting to see some mind-blowing technology emerge. The latest research in turning low-resolution photos into high-definition photos may drop your jaws — it’s starting to cross into the realm of sci-fi.
What you see above isn’t a mockup of what these enhancement technologies may one day do: it’s an actual example of an AI-powered enhancement done by a new algorithm called EnhanceNet-PAT developed by scientists over at the Max Planck Institute for Intelligent Systems in Tübingen.
The example started with this original photo of a bird:
The scientists took the photo and created this low-resolution version in which all of the fine details are lost:
The low-resolution version was then processed by EnhanceNet-PAT, creating an artificially enhanced high-definition version that is virtually indistinguishable from the original photo.
What the fine details look like are obviously different in the enhanced version compared to the original one, but what’s lost in accuracy is made up for in realism.
“While the pixel-perfectness is being sacrificed, the reward is a better result,” the Max Planck Institute writes.
Here are some more examples of low-res photos (on left) and the enhanced high-res versions produced by the algorithm (on right):
Creating large photos from small ones is a field known as single image super-resolution (SISR) technology. Yesterday, we featured a free website called Let’s Enhance that uses one SISR strategy to offer this type of image enhancement as a service.
Traditional upscaling technologies have tried to fill in missing pixels and details by calculating looks based on surrounding pixels. The results of these types of strategies are blurry and unsatisfactory. What scientists are now exploring is using machine learning to “learn” what features low-resolution photos should have by also studying the original high-resolution versions.
Once trained in this way, the algorithms are able to take new low-res photos and make a best guess for what an “original” high-res version of that photo would look like.
“By being capable of detecting and generating patterns in a low resolution image and of applying these patterns in the upsampling process, EnhanceNet-PAT thinks how the bird´s feathers should look like and adds extra pixels to the low-resolution image accordingly,” the Max Planck Institute says. “You could say the technology created its own reality. For most viewers, the result is very much like the original photo. The picture of the bird is good to adorn the photo album.”
from PetaPixel http://ift.tt/2ilkG4S
from WordPress http://ift.tt/2z6DlcI