Fuzzy photos can be clarified by Google AI
Artificial intelligence systems built by Google translate images that are heavily pixelated and of poor quality into clear human or object photos.
Computer scientists at Google Brain from the Google AI Center team have shown that this not only enhances the resolution of the picture but also fills in the missing details of the picture.
The combined analysis of the adjusted neural network and the previous neural network can produce a higher resolution 32 × 32 pixel version. From the process can see a fuzzy, almost unrecognizable picture into a clear person or thing.
Source: wired.co.uk
AI system to take a two-pronged approach. Adjusting the network The nervous system uses a low-resolution image and compares it to a high-resolution image to determine if there are any people or things in the image. The researchers explained that low-resolution images and large images can be scaled down to the same 8×8 pixel size for comparison with high-resolution images.
Google AI researchers wrote in their essay: “When the source images are not complete, then their challenge is not just to blur the image but to create parts of the image that are missing.”
To prove that AI-generated images are credible, the researchers tested their systems on human volunteers. They showed the source image and the generated image to a group of volunteers, allowing them to see which one is the source image. Fifty percent of people view synthesized images as source images, and one algorithm confuses people’s vision.
Source: wired
