The party is ON! Join us at TNW Conference 2021 in Amsterdam for face-to-face business!

The heart of tech

This article was published on October 31, 2017

    This new AI can make your low resolution photos great again

    This new AI can make your low resolution photos great again
    Camille Charluet
    Story by

    Camille Charluet

    Former Editor, TNW Spaces

    It’s extremely aggravating when you decide to print a picture you took on your phone, only to find it’s blurry, grainy or pixelated. No matter how much of a photoshop master you might be, those low-resolution photos are nearly impossible to enhance.

    Thankfully, scientists have discovered a new way to create high-resolution versions of low-grade photos using artificial intelligence – and the results are quite impressive.

    Developed out of the Max Planck Institute for Intelligent Systems in Germany, researchers propose a new approach to traditional (and usually disappointing) single-image super-resolution (SISR) technology on the market.

    Usually the last resort when photoshop fails, the typical SISR software attempts pixel-perfect reconstruction by adding extra pixels and averaging them with surrounding pixels on blown-up images. Unfortunately, the results are commonly blurry images that lack sharpness like the below:

    Credit: PetaPixel

    EnhanceNet-PAT is the German researchers new AI-driven alternative. By using machine learning, their software aims for faithful texture synthesis rather than perfection. Scientist Mehdi M.S. Sajjadi from the Max Planck Institute for Intelligent Systems explained how their new technique works:

    The algorithm is presented millions of low-resolution images, which he is to scale up. He tries on the task and then he is presented for comparison each of the high-resolution original photo. The algorithm perceives the difference and learns from its mistake.

    While the results are not a true match to the original, EnhanceNet-PAT imagines a high-resolution image and adds pixels to the low-resolution image accordingly. Practice makes perfect and much like a human, once the software is trained it no longer needs to be fed original photos. See the results for yourself:

    Credit: MPI-IS
    EnhanceNet-PAT upsampled a low-resolution image (left) to a high definition version (middle). The upsampled version is quite similar to the original image (right). Credit: Max Planck Institute for Intelligent Systems

    Thanks to EnhanceNET-PAT, you can now stop cursing at that frighteningly blurry holiday photo you took on your phone and decided to print. The software is finally here to make it work and the results look promising.