High Scale Image Resizing Now Possible With Google Brain

Last month, Google released its new software as part of the Google Brain project, that can create a clearer and more detailed image from extremely small or pixelated images.


We have all been there – taken a picture on your phone camera only to find out that the image ended up being blurry. Instead of showing a beautiful picture, you are left with a highly pixelated image that just isn’t worthy of sharing with anyone. Well, in that case, Google now has a solution for you.

The software is being regarded as a groundbreaking software in image recognition and can help in numerous fields such as investigations and picture recognition regarding detecting faces from CCTVs and Road Traffic Cameras.

However, even Google can’t do the unthinkable and produce an image out of something that doesn’t exist. But what they have managed to do is create a clearer image out of a tiny pixelated one with the use of two highly advanced neural networks.

The first component is called the Conditioning Network, which uses multiple high-resolution images in comparison to the small pixelated images to find a match. The high-resolution image is matched by scaling it down to an 8×8 picture, making it similar to the small pixelated image.

The second component of the network uses a software called Pixel CNN which is the part responsible for actually making the look clearer. The software adds realistic details to the image to make it look more realistic. Google Brain works by combining these to software to create a picture which can be enhanced to have a higher resolution and also less blurry. For example, if there is a black pixel pattern on an image, and it matches a particular pixel pattern of the other picture, which is, let’s say the hair of the person, the smaller image will get the details of the hair filled in to show a clearer picture.

However, even with this advanced technology in place, Google Brain still cannot make an image that is identical to the original image, but a relatively close match. When Google decided to test out this software, the results were incredibly resourceful in helping develop high-resolution images which otherwise were almost indiscernible. It is, however, important to know that the image created may have some discrepancies which may result in the picture not adhering to every single facial detail of the actual person. This can lead to a few problems when it comes to using this technology to look at surveillance footage during the investigation or when conducting forensic analysis. The software merely adds details to the original image to make it seem clearer. The software acts as a more advanced image development tool if anything. However, in some cases, it can help give police adequate information by narrowing down on a particular physical type while trying to catch suspects to help their investigations.

If you have ever been one to watch crime detective shows, you may have seen numerous investigators saying things like ‘enhance the reflection’ or ‘zoom in on that image.’ With Google Brain, this picture recognition isn’t just a cool thing they say on television. It’s now slowly going to become a part of everyday life.

Related posts

Leave a Comment