You know how in CSI, the cops always try to “enhance” a shot to zoom in and read (non-existent) details in photos? It’s amusing to the rest of us, but perhaps one day won’t be all that impossible, with artificial intelligence.
Researchers have been adopting neural networks and machine learning technologies to help computers fill in missing detail in photos.
Some consumer-ready websites are already making some of this magic accessible to you and me.
One of these is called Let’s Enhance, which was recently launched with an easy drag-and-drop interface.
After you let its system process your picture, you’ll be able to download a shot that’s been enlarged by four times, by clicking the “magic” option.
The results are certainly impressive. We compared it with several small pictures that were enlarged with a basic Photoshop interpolation method.
Looking at this shot of people’s hands, Let’s Enhance has restored the wrinkles in the texture of the hand in the foreground.
It may have been a little too enthusiastic with this, producing fairly harsh outlines, but it’s certainly reproduced the original picture quite faithfully.
And if you look at the cat above, you can see the jagged lines around its eyeballs, where basic interpolation has simply filled in the missing pixels by copying the ones around them.
Let’s Enhance’s version has restored the smoothness of the lines, and even reproduced the reflection of the photographer in the blacks of the cat’s eyes.
SRGAN, for example, relies on a generative adversarial network, which basically pits two neural networks against each other to test and fine-tune an image repeatedly.
Looks like some of our TV magic is coming to the real world one day. Enhance!