New AI system can make low-resolution photos clear

When you decide to print photos you took with your phone, you often find that they turn out blurry, and sometimes even Photoshop doesn’t help. Luckily, scientists have discovered a new way to turn low-resolution photos into high-resolution versions using artificial intelligence — and the results are impressive.

Scientists at the Max Planck Institute for Intelligent Systems in Germany have developed an AI-based single image super-resolution (SISR) reconstruction technique.

Processing the image with typical SISR software usually results in a blurry, unclear image like this:

The AI-based EnhanceNet-PAT uses machine learning, and its software is designed to accurately and reliably synthesize textures. Mehdi MS Sajjadi, a scientist at the institute, explained that they fed millions of low-resolution images to the AI ​​system. The algorithm was then used to try the task, and the researchers provided a high-resolution version of each photo. The algorithm sensed the differences and learned from its mistakes.

Although the results are not consistent with the original ones, EnhanceNet-PAT can imagine the high-resolution image and add pixels to the low-resolution image accordingly. Once the software is trained, the researchers no longer need to input the original photos.