Google can now identify heavily pixellated images, thanks to the Google’s AI which is a combination of two artificial intelligence Google Brain neural network systems built by Google. Heavily pixellated images are the low quality images. With Google’s AI, these can be enhanced into a clear photo of a person or object. This artificial intelligence system, developed by central Google AI Team, has shown that the system not just enhances picture resolution but also fills in the missing details and teaches itself to create its own encryption.
Three researchers from Silicon Valley, in the paper – Pixel Recursive Super Resolution, trained their system on 8×8 pixel images of celebrities and photos of bedrooms. Then a combination of conditioning and prior neural network analyses and produces 32×32 pixel versions of the same image. This turns the blurry and almost unrecognizable image into an image which clearly represents a human or object or room or whatever the picture was about.
The AI Systems work with a two-pronged approach. The conditioning network compares the low quality image with high-resolution images to determine whether the image has a face or a room or some other thing in it. This is done by scaling down the high-resolution image to 8×8 pixel size.
“When some details do not exist in the source image, the challenge lies not only in ‘deblurring’ an image, but also in generating new image details that appear plausible to a human observer. By incorporating the prior knowledge of the faces and their typical variations, an artist is able to paint believable details.“
– the Google Brain researchers wrote in their paper.
When both images are of the same pixel size, it is comparatively more simpler for the AI to identify similar pixels and shapes among the two versions. Once the conditioning AI network has completed its job, PixelCNN is used to add extra pixels to the 8×8 image by figuring out what it already knows about images of that kind. At the end of each neural network process, the results are combined to create a final image.
This method will surely be of a big benefit in the future. One of them could be that it would enable us to see clearly what is recorded by the blurry CCTV cameras. Further development is going on under this which will help in converting a low quality video to a high quality video. So if something is recorded, rest assured that any person in the video or image can be identified with ease. At present, some more tests are needed on this system though.