[NEW TECHNOLOGY] Scientists Developed AI That Makes Blurry Pictures Into Hyperrealistic Photos: 60x More Sharper

Scientists have developed a new artificial intelligence that can blurry pictures into hyperrealistic photos--it is 60 times sharper than the original one. According to Tech Xplore's latest report, the AI tool that can transform unrecognizable, blurry photos of people's face into eerily convincing computer-generated portraits and more detailed pictures, was developed by the researchers of Duke University.

The report stated that an image of a face could be scaled up by previous methods only up to eight times the original image. However, a new way was discovered by the Duke team to take a handful of pixels, creating realistic-looking faces that are 64 times the resolution, making the features such as stubble, fine lines, and eyelashes, which cannot be seen in the original image, visible in the new photo.

"Never have super-resolution images been created at this resolution before with this much detail," said Cynthia Rudin, a Duke computer scientist who led the researchers.

The researchers clarified that the new AI tool doesn't have the capability to make an out-of-focus, unrecognizable image from a security camera into a high-resolution photo of a real person. But, it can generate new faces, which doesn't really exist, that is plausibly realistic.

New AI tool that makes blurry photos into hyperrealistic pictures, 60x sharper, developed by scientists

The report stated that although the AI tool's concept is focused by the scientists in faces, it can also be applied in medicine and microscopy, also in satellite and astronomy imagery to take low-resolution shots of almost anything and transforming into realistic-looking, sharp pictures.

According to Duke Today's latest report, a tool in machine learning called GAN or "generative adversarial network," two neural networks trained on the same data set of photos, was used in the study conducted by the Duke scientists. The first network can mimic the original images it was trained on, developing AI-created human faces while the second network is responsible for identifying if the output is convincing enough to be mistaken for the original photo.

Researchers claimed that the first network keeps getting better and better as the experiment continues, making the second network unable to distinguish between the original image and the AI-created image. The researchers' newly discovered method, which can create realistic images from noisy, poor-quality input, is called PULSE. Rudin said that the AI tool can create any number of uncannily lifelike possibilities from a single blurred image of a face, each of which looks slightly different from the original face.

Also Read: [New Discovery] Unusual Echoes at the Center of the Earth Seen and Heard by Scientists: What Is It?