Google is becoming smarter and smarter as the days are passing and they are implementing Artificial Intelligence at an immense pace.
With Google Artificial Intelligence systems, Google will be able to convert a heavily pixelated low-quality image into a clear photo so that it can further be used for identification or other purposes.
This makes the use of Artificial Intelligence a must have for identifying criminals or other such activities.
Some computer scientists from Google, the central Google AI team have shown in the past that they can not only enhance a picture quality but they can also fill the gaps that are present in the picture.
This becomes useful in circumstances where you have a photo of a verdict but you are not able to identify the whole picture of that person then you can use this system to make sure that the right person is getting caught.
Some researchers have stated in a paper Pixel Recursive Super Resolution that they have trained 8*8 pixel images of celebrity faces and photos of bedrooms.
Here the conditioning neural network and a prior neural network is combined so that it can produce higher resolution of 32*32 pixel versions of the image.
In a nutshell, you will be able to convert a blurry, almost unrecognizable picture into a one that is clearly represented by a human.
Let us break the system into steps.
What happens is that the low resolution images are converted into high resolution images by scaling down the low resolution large image to the same 8*8 pixel size.
The Google researchers have written in the paper that the challenge is not only describing a process on How to unblur a picture but also in adding new details so that the image look feasible to a human observer.
First both images are compared to each other. Both of them are of the same sizes and it is easier for them to similar pixels and shapes between both the versions.
As an example, both the systems can recognize ears that resemble a particular shape and compare all of them by comparing the pixels in the other image.
In addition, the Google Researchers use the PixelCNN so that it can add the pixels that are extra to the 8*8 image.
Now here there is a thing that needs to be noted and that is if lips are in pink color, then pink pixels are to be identified separately.
Google’s AI just created its own universal ‘language’
As every neural network process ends the Google researches sums up and gives us the result so that we can create a final image.
In order to prove that the AI generated images are believable, the researchers tested the system on human volunteers.
A group of humans were shown a low-quality image and the other group of people was shown an image that was created by AI. Both groups were asked to check which photo was taken from a camera.
Here you would like to note that 10 percent of the people thought that the Artificially created image was taken from a camera. This was a plus point.
In the near future, what you can do is innovate this process and add detail to the low-resolution images and videos.
An application to this thing is blurry CCTV images.
Here you would like to note is that this method needs to be tested with a collection of images.