Edge browser will enhance all web images with AI


Microsoft Bing has announced new artificial intelligence technology that will bring a 4K image experience to websites through Microsoft Edge, automatically enhancing website images. The technology, called Turing Image Super-Resolution, allows images to be displayed at high resolution, regardless of the quality of the original image.

The new technology was developed by Microsoft Turing AI Project Development Team.

Already used in Bing Maps

The new technology is already being used in Bing Maps to improve the quality of their satellite aerial imagery.

Below is an aerial image comparison of Google’s headquarters in Mountain View, California.

The Bing Maps screenshot is on the left and the corresponding Google Maps image is on the right:

Bing Maps vs. Google Maps

Side-by-side comparison between Bing Maps and Google Maps Aerial Images

How Microsoft built the technology

Four important ideas led to the success of the model.

  1. Human reviewers
  2. Noise modeling
  3. Loss of perception and GAN
  4. Transformers for Vision: Enhance and Zoom

Human reviewers

Microsoft realized that the metrics used to measure the success of image-bound models did not match human visual perception. So they created a side-by-side visual comparison tool that used human raters to help gauge the success of the model.

Noise modeling

Microsoft has taken the approach of starting with high quality images, then degrading them by adding noise to them, then teaching the model to bring the image back to the original high quality state of the image .

Loss of perception and GAN

This was part of efforts to align the results with the human vision.

Microsoft’s announcement said:

“…we found that optimizing our models using only the pixel loss between the output images and the ground truth images was not sufficient to produce the optimal output aligned with the perception of the human eye.

In response, we also introduced perceptual loss and GAN and set an optimal weighted combination of the three losses as an objective function.

Transformers for vision

Microsoft leveraged the power of transformers used in language models, focusing on enhancement and zooming.

This means improving the image and also focusing on scaling the image, which is a hard thing to do.

Generally, it is easy to reduce an image. But taking a small image and scaling it up usually ends up magnifying the low-resolution artifacts of the original image.

The researchers therefore created a system capable of calculating and “recovering” the missing image data from the lower resolution image and bringing it to a higher resolution.

Microsoft calls the process of scaling an image, DeepZoom.

Edge: 4K TV from web browsers

Microsoft sees this new AI feature as a way to bring a 4K visual experience to web browsing, as well as improve video meetings and family photos uploaded to the web.

The technology is already available in the experimental version of Edge called Canary Edge.

The new feature will be rolling out to the consumer version of the Edge browser over the next few months.

Quote

Read Microsoft’s announcement

Turing image super-resolution

Leave a Comment

Your email address will not be published.