Scientists have evolved an artificial intelligence AI system that can create a high-definition version of a low-resolution image. Technology to produce a large-sized image from a low-resolution image is known as single-image super-resolution (SISR) technology.
SISR has been reportedly studied for decades, but with limited results. The software connects extra pixels and averages them with the surrounding pixels, but the result is blurriness.
The researchers at the Max Planck Institute for Intelligent Systems in Germany suggested a new approach giving images a realistic texture when magnified from small to large using machine learning.
They wanted to expand previous image generation event including efforts by Google, by both creating higher-quality images and generating a wider variety of computer-generated images in less time. To do that, the researchers created a progressive system. Since AI learns more when data is fed into the system, the group added more difficult renderings as the system progressively improved.
The program launched with generating low-resolution images of people that don’t actually exist, motivated by all the photos in the database, which are all images of celebrities. As the system improved, the researchers added more layers to the program, adding more fine detail into low-resolution images became 1080p HD standard photos. The result is high-resolution, detailed images of “celebrities” that don’t actually exist in real life.
The team applied artificial intelligence and an adaptive algorithm for upsampling the image learns from experience to improve the result. The learning process is much like that of a human, researchers said. “The algorithm is given the task of upsampling millions of low-resolution images to a high-resolution version, and is then shown the original,” said Mehdi MS Sajjadi from Max Planck Institute of Intelligent Systems.
Researchers evolved the EnhanceNet-PAT technology that once trained, no longer needs the original photos. The technology is more efficient than any other SISR technology currently on the market. In contrast to existing algorithms, Enhance Net-PAT does not attempt a pixel-perfect reconstruction, but rather aims for faithful texture synthesis, researchers said.
By uncovering and generating patterns in a low-resolution image and using these patterns in the upsampling process, Enhance Net-PAT adds extra pixels to the low-resolution image accordingly, they said. For most viewers, the result is very similar the original photo, researchers added.